1. I The similarities and systematic differences between English …web.khu.ac.kr/~jongbok/research/final-papers/kim-sag... · 2015-03-27 · our proposed analysis. 2.1.1. Lexicon

JONG-BOK KIM and IVAN A. SAG

NEGATION WITHOUT HEAD MOVEMENT�

This paper presents a lexicalist analysis of negation in French and English. In bothlanguages, negation in finite clauses is grammatically distinguished from constituentnegation. Lexical idiosyncrasy motivates treating finite negation as a verbal comple-ment, while constituent negation is treated in terms of a familiar modifier-head con-struction. General principles ordering lexical and phrasal heads ensure that negation(the adverbs not and pas) follows the finite verb (the finite auxiliary verb in English),while only constituent negation appears preverbally. Our constraint-based account,cast within the framework of Head-driven Phrase Structure Grammar (HPSG), pro-vides a viable alternative (with broader coverage, fewer devices and simpler princi-ples) to analyses based on head movement, which seek to explain the syntax of nega-tion and adverbial positions in terms of the interaction of morphological properties,verb movement, and functional projections.

1. INTRODUCTION

The similarities and systematic differences between English and Frenchgrammar have received considerable attention in the recent syntactic liter-ature. Central to this inquiry has been the following set of contrasts:

� Some of the material contained in this paper was presented at the 14th West CoastConference on Formal Linguistics (University of Southern California: March 10–12, 1995)and is published as Kim and Sag 1995. We are particularly grateful to Daniele Godard andPaul Hirschbuhler for detailed discussions and valuable insights on the French materialpresented here. We also thank Anne Abeille, Bob Borsley, Chan Chung, Peter Culicover,Tony Davis, Tibor Kiss, Marie Labelle, Bob Levine, Rob Malouf, Mike Maxwell, PhilipMiller, Fritz Newmeyer, Byung-Soo Park, Carl Pollard, Christine Poulin, Peter Sells, Hen-riette de Swart, Elizabeth Traugott, and Tom Wasow for valuable comments, suggestions,and help of various kinds. Three anonymous reviewers also deserve our special thanks fortheir critical comments that helped us reshape this paper. We also thank three other review-ers for their suggestions, which improved the revision. Of course, we alone are responsiblefor the views reflected in this paper and for any errors that might remain. This research wassupported by the Korea Research Foundation (grant number KRF-99-042-A00007), with-out whose generous financial support the work reported here would not have been possible.This research was also partly conducted in conjunction with CSLI’s Linguistic GrammarsOnline (LINGO) project. In that connection, we gratefully acknowledge the support of theNational Science Foundation (grant number IRI-9612682).

2 JONG-BOK KIM AND IVAN A. SAG

Position of Negation:

(1) a. *Kim likes not Lee.

b. Kim does not like Lee.

(2) a. Robin n’aime pas Stacey.

Robin (n’)likes NEG Stacey

‘Robin does not like Stacey.’

b.*Robin ne pas aime Stacey.

Position of Adverbs:

(3) a. *Kim kisses often Lee.

b. Kim often kisses Lee.

(4) a. Robin embrasse souvent Stacey.

b. *Robin souvent embrasse Stacey.

Subject-Verb Inversion in Questions:

(5) a. *Likes he Sandy?

b. Does he like Sandy?

(6) a. *Likes Lou Sandy?

b. Aime-t-il Sandy?

Drawing on the earlier insights of Emonds (1978), Pollock (1989), and anumber of subsequent researchers (Belletti (1990), Zanuttini (1991, 1997),Chomsky (1991, 1993), Lasnik (1995), Vikner (1997), and Haegeman(1997), inter alia) have interpreted these contrasts as providing criticalmotivation for the process of head movement and the existence of func-tional categories such as MoodP, TP, AgrP, and NegP. It has been widelyaccepted that the variation between French and English illustrated here canbe explained only in terms of the respective properties of verb movementand its interaction with a view of clause structure organized around func-tional projections.

For example, in Pollock’s (1989) proposal, all verbs in French move toa higher structural position, whereas this is possible in English only for theauxiliaries have and be, as shown in (7):

NEGATION WITHOUT HEAD MOVEMENT 3

(7) a. French:TP

Tns NegP

pas Neg�

Neg AgrP

ne Agr VP

t V ...

all verbs

b. English:

TP

Tns NegP

Neg AgrP

not Agr VP

t V ...

have/be

Why does V-movement happen when it does? This question has beenanswered in diverse (and sometimes inconsistent) ways in the literature (cf.Pollock 1989, 1994, 1997a, and 1997b, Vikner 1997). In Pollock 1989, it isthe strength of the Agr feature that determines V-movement: unlike French,English non-auxiliary verbs cannot undergo V-movement because Agr inFrench is ‘transparent’ (or ‘strong’) whereas Agr in English is ‘opaque’ (or‘weak’). The richness of French verbal morphology is assumed to providethe motivation for the strength of French Agr, in consequence of whichthe raised verb in French can transmit theta roles to its arguments throughAgr, thus avoiding any violation of the theta criterion. But the weaknessof English Agr (assumed to follow from the paucity of English verbalmorphology) is what blocks lexical verbs from assigning theta roles oncethey have moved to Tns. Hence movement of a theta-assigning verb in


English would result in a violation of the theta criterion. The basic spiritof this analysis—that ‘morphology determines syntactic movement’—hasremained essentially unchanged in subsequent research (Pollock 1997a,1997b, Chomsky 1995) though what triggers V-movement has varied con-siderably in subsequent work.� As far as we are aware, there is no agreedupon movement-based analysis of either the English or French systems. Infact, as Lasnik (2000: 181–190) stresses, the Minimalist Program as artic-ulated in Chomsky (1993) not only fails to deal with the ungrammaticalityof simple examples like *John left not or *John not left, it also provides nobasis for explaining the French/English contrasts in adverb position dis-cussed by Pollock (e.g. embrasse souvent vs. often kisses).

In this paper, we offer a radically different perspective on the gram-mar of negation and related phenomena. The surface-based, lexicalist treat-ment we develop—which makes no appeal to head movement or any othermovement operation—provides a detailed, transparent account of negationin each language and allows us to elucidate the similarities and systemat-ic differences between the two systems. We will limit our attention hereto French and English (as did Pollock (1989)), though in so doing it willno doubt seem to some readers that we have ignored massive independentcross-linguistic evidence for head movement. So vast, in fact, is the lit-erature based on head movement and functional projections, that by nowit appears to many that those analytic constructs have been firmly estab-lished on empirical grounds. But such appearances can be deceptive, aswe will attempt to show. Though space requirements dictate that we care-fully delimit the scope of the present inquiry, we hope that our results willlead others to rethink the transformational orthodoxy that underlies con-ventional approaches to related phenomena in other languages as well.

2. THEORETICAL FOUNDATIONS

2.1. Overview

A descriptively adequate theory of grammar must include at least the fol-lowing components:

� In Pollock 1997a, V-movement is driven by ‘mood’ distinctions, where modals, haveand be are interpreted as mood markers. In Pollock 1997b, by contrast, V-movement isdependent upon ‘interpretable’ or ‘uninterpretable’ ‘person’ features. From Haegeman’s(1997) perspective, English and French are both claimed to have V-movement. The dif-ference between the two languages comes from the fact that French verbs move to I atS-structure and English finite lexical verbs move to I at LF. The English verbs can waituntil LF because their features are ‘interpretable’.


(8) a. A system of lexical representation.

b. Principles specifying the nature of headed phrases and their rela-tion to those lexical representations.

c. Principles of linear order.

d. A ‘root clause’ definition.

There are many competing accounts of (8)a, of course, all of which seekto organize lexical representations (minimally) in terms of part of speechinformation and constraints on argument structure and argument selection.We will assume one such organization here, but others are possible. Themost common assumptions about (8)b include a version of X-Bar Theory,augmented by some mechanism requiring that the sisters of a lexical headbe categorially nondistinct from the appropriate arguments selected by thathead. Linear order is usually derived from headedness parameters (e.g.head first) and their interactions with other grammatical constructs, suchas transformational movement. The definition in (8)d, analogous to thestipulation of an ‘initial symbol’ in a context-free grammar, usually goesunmentioned in the literature, but some such principle must be included inany complete grammar in order to guarantee, for example, that only finitesentential structures (e.g. trees rooted in TP[+fin]) may function as root (orindependent, ‘stand-alone’) clauses.

The theory of grammar we advocate is truly compact, in that it includeslittle or nothing more than a precise specification of the devices in (8).� Ourtheoretical point of departure is HPSG, roughly as presented by Ginzburgand Sag (2000) or (in simplified form) in Sag and Wasow 1999. We beginby reviewing the particular theoretical constructs that will be relevant toour proposed analysis.

2.1.1. LexiconIn HPSG, lexemes—in fact, all linguistic objects—are organized in termsof classes, or types. The hierarchical organization of linguistic types allowsconstraints to be stated over various natural classes of words (those thatcorrespond to superordinate types). Lexical entries are partial descriptionsof lexemes that provide only the information that is not given by moregeneral constraints on the lexeme’s maximal (most-specific) type or by oneof its supertypes. A lexical entry of course includes syntactic, semantic

� Transformational theories, by contrast, including (somewhat paradoxically) those thatfly a ‘minimalist’ banner, embrace the view that grammatical theory includes much more.For relevant critical discussion, see Johnson and Lappin 1999.


and phonological constraints, but for present purposes we will focus onrelevant syntactic information specified in terms of valence features andthe feature HEAD. The combination of lexeme-specific and more generaltype constraints yields a lexical entry for a nonauxiliary English verballexeme like love that includes the information shown in (9):

(9) love��

v-lxm

HEAD

�verb

AUX �

�

SUBJ h NP i

COMPS h NP[acc] i

��

Note that here the value of the feature HEAD, identifying inherent prop-erties of a given lexeme, is itself a feature structure (of type verb) thatincludes various feature specfications that will be ‘passed up’ from headdaughter to mother in a headed construction, according to the Head FeaturePrinciple discussed below.

Verbal lexemes like (9) give rise to (inflected) words. Word formationcan be thought of in terms of applying an inflectional rule to a lexemedescription to produce a word description or (equivalently, for presentpurposes) as building a word from a lexeme via a non-branching lexi-cal construction.� The lexeme in (9) will give rise to nine distinct words,assuming that we distinguish 3rd-singular-present-indicative, non-3rd-singular-present-indicative, past-indicative, subjunctive/imperative, base,present-participle, gerund, perfect-participle, and passive-participle. The3rd-singular-present-indicative form is sketched in (10):

(10) loves��

word

HEAD

��verb

AUX �

FORM fin

��

SUBJ h NP[nom,3rd,sg] i

COMPS h NP[acc] i

��

Here the SUBJ and COMPS features specify constraints on the number andkind of valents that loves must combine with. In addition, finiteness islexically encoded as a projected lexical property.

� For more on different approaches to expressing lexical regularities in HPSG, seeMeurers 2000, Koenig 1999, Copestake and Briscoe 1992, Bouma et al. 2000 and (fora more elementary presentation) Sag and Wasow 1999.


Given this rich encoding of lexical information, HPSG is able to incor-porate the principle of Strong Lexicalism (Scalise 1984: 101ff) and theclosely related Lexical Integrity Hypothesis of Bresnan and Mchombo(1995). That is, the principles of word formation are independent fromthose governing syntax and internal word structure and morphological ele-ments and morphological structure are invisible to syntactic constraintsand operations.

2.1.2. Phrasal ConstructionIn the version of HPSG assumed here, a grammar includes an inventoryof construction types. This organization serves to organize constraints onphrases: each maximally specific type of phrase inherits constraints fromits superordinate types. These constraints impose restrictions on the motherof the construction and on its daughters; hence together they function muchlike a rewrite rule in a standard context-free grammar. But the constructiontypes, some of which are universal (see Ackerman and Webelhuth’s (1998)discussion of ‘archetypes’), are organized into a cross-classifying multidi-mensional hierarchy, so that ‘family resemblance’ properties of construc-tions can be expressed as generalizations: constraints on a common super-type. Ginzburg and Sag (2000) offer a two dimensional classification ofphrasal construction types, a portion of which is illustrated in (11):�

� Here and throughout, we use the following abbreviations: core-cl (core-clause),decl-cl (declarative-clause), inter-cl (interrogative-clause), excl-cl (exclamative-clause),rel-cl (relative-clause), decl-hd-su-cl (declarative-head-subject-clause), wh-int-cl (wh-interrogative-clause), wh-excl-cl (wh-exclamative-clause), hd-ph (headed-phrase), hd-adj-ph (head-adjunct-phrase), hd-comp-ph (head-complement-phrase), hd-subj-ph (head-subject-phrase), hd-fill-ph (head-filler-phrase), non-hd-ph (non-headed-phrase).


(11) phrase

CLAUSALITY

clause

rel-cl core-cl

decl-cl

decl-hd-su-cl

inter-cl

wh-int-cl

excl-cl

wh-excl-cl

non-clause

HEADEDNESS

hd-ph

hd-adj-ph hd-comp-ph hd-subj-ph hd-fill-ph

non-hd-ph

A type-based classification allows some generalizations to be stated invery general terms (e.g. as a constraint on the type phrase), in maximallyspecific terms (e.g. as a constraint on a maximal construction type), or asholding over a particular family of constructions (e.g. as a constraint onan intermediate-level phrasal type such hd-adj-ph, hd-comp-ph, or inter-cl). As noted by Ginzburg and Sag (2000), there is considerable evidencethat this is precisely the way natural languages are organized, in terms ofconstraints that characterize diverse, cross-cutting classificatory grains.

As shown in (11), each maximal phrase type is cross-classified, inher-iting both from a CLAUSALITY type and from a HEADEDNESS type. Thatis, each type of phrase is classified in one dimension as a particular kindof clause or else as a nonclause. At the same time, each type of phraseis classified as an unheaded� phrase or else as a particular kind of head-

� Coordination constructions are one example of an unheaded phrase type. See Pollardand Sag 1994 and Borsley 1994.


ed phrase (head-subject phrase, head-complement phrase, etc.) With thephrasal multiple-inheritance hierarchy, the theory need not posit phan-tom formatives, such as the inaudible functional heads that are assumedin many competing analysis of clausal structure (see Ginzburg and Sag2000 for discussion). The work done by these elements is replaced by con-straints associated with the various types of clause. For example, phrasesare subject to the following general constraints:

(12) a. The Generalized Head Feature Principle (GHFP), which—bydefault—identifies the value of all features of the mother with thatof the head daughter in all headed phrases.

b. The Empty COMPS Constraint (ECC), stating that all phrases havethe empty list as the value of the feature COMPS.

c. Valence constraints on particular kinds of headed phrase. Theseidentify the selectional properties of the head daughter (i.e. thehead daughter’s values for the feature SUBJ or COMPS) with theappropriate non-head daughter.

The GHFP, formulated in (13), simply enforces the identity between‘X’ and ‘XP’ that is familiar from X-theory:�

(13) Generalized Head Feature Principle:

hd-ph:

[SYNSEM / � ] � ... H[SYNSEM / � ] ...

Since the GHFP governs all headed constructions, it also applies to head-complement constructions, which are further constrained as shown in (14):

(14) hd-comp-ph:

[ ] � H

�word

COMPS A � list

�, A nelist

� Here we follow the standard practice of using boxed integers (or ‘tags’) to indicatefeature structure identities. Note that boxed capital letters are used to identify feature valuesthat are not feature structures, but rather lists of feature structures. We adopt the theory ofdefaults (and the ‘/’ notation for default values used in (13)) outlined in Lascarides andCopestake 1999.


This construction permits a lexical head daughter to combine with com-plements that it selects via the COMPS feature.� Hence complements areintroduced as sisters of the lexical head, i.e. lower in a structure than eithera subject or a specifier. The ECC in (12)b guarantees that the COMPS valueof any phrase is the empty list. This interacts with the valence constraintsmentioned in (12)c to ensure that the head daughter’s SUBJ and COMPS

values are either ‘cancelled off’ or else simply inherited by the mother in aheaded phrase. For example, the ECC interacts with the constraints in (14)to ensure that a hd-comp-ph’s COMPS value is the head daughter’s COMPS

list ‘minus’ whatever complements are realized inside the hd-comp-ph (orelse left unexpressed in ellipsis). By contrast, since the constraints in (14)say nothing about SUBJ, the GHFP guarantees that the hd-comp-ph’s SUBJ

value is identical to that of its head daughter. In the case of verb phrases,this value will be a list whose single member corresponds to the subject ofthat VP.

Principles like those in (12)b,c thus replace the notion of ‘bar-level’with a notion of ‘cancellation of arguments’, familiar from work in cate-gorial grammar. The result is a somewhat unusual ‘barless’ X-theory.

2.1.3. Stand-Alone PhrasesTo specify which clauses can function as independent utterances, we alsoposit a principle stating simply that stand-alone (or ‘root’) phrases must beconsistent with the information in (15).�

(15)��

phrase

HEAD

�verb

FORM fin

�

SUBJ h i

...

��

Of course, our account must ultimately be extended to allow other kinds ofutterances, including fragments, short answers to questions, interjections,and so forth, but we will ignore such complications here.

� The constraint in (14) says that a hd-comp-ph must contain a non-empty list (nelist)of complement daughters and that these complements must correspond to an initial sublistof the head daughter’s COMPS value. The type fin-vp discussed below (a subtype of hd-comp-ph) imposes the further requirement that all selected complements be realized, butthe elliptical verb phrase construction discussed by Sag (to appear) explicitly requires thatsome complement selected by the head not be realized within the VP.

� This constraint is incomplete in various ways that are explained more fully inGinzburg and Sag 2000 and Sag and Wasow 1999. Finite clauses satisfying the descrip-tion in (15) can of course also appear in appropriate embedded environments.


2.2. Two Construction Types of English

Our system of constraint inheritance accounts for the shared properties ofindividual constructions in a principled fashion. For example, a VP likelikes Sandy is an instance of the type finite-verb-phrase (fin-vp), whoseplace in the construction hierarchy is shown in (16):

(16) phrase

CLAUSALITY

non-clause

HEADEDNESS

hd-ph

hd-comp-ph

fin-vp

likes Sandy

Because fin-vp is a subtype of hd-comp-ph, which in turn is a subtypeof hd-ph, it follows that an instance of the type fin-vp must simultaneouslysatisfy the more general constraints on those supertypes, as well as theparticular constraint stated in (17):

(17) fin-vp (preliminary version):

[ ] � H

�FORM fin

COMPS A

�, A

The effect of the various constraints placed on finite verb phrases is illus-trated in (18):

We discuss the motivation for distinguishing finite VPs from other head-complementconstructions in section 7 below.


(18) VP��

fin-vp

HEAD �

��verb

AUX �

FORM fin

��

SUBJ B h NP[nom,3rd,sg] i

COMPS h i

��

H

V��

HEAD �

SUBJ B

COMPS h � i

��

� NP��HEAD

��CASE acc

PER 3rd

NUM sg

��

likes Sandy

As (18) shows, the GHFP causes the lexically encoded HEAD informationdiscussed above to be ‘passed up’ from the lexical head daughter to themother. Other information (e.g. the SUBJ value) is passed up from the lex-ical head as well, unless some constraint says otherwise. The constraint in(14), for example forces the head daughter’s COMPS value to be nonemp-ty, while the ECC in (12b) forces the mother’s COMPS value to be empty,thus overriding the GHFP in this one case: mother and head daughter musthave distinct COMPS values. In addition to the GHFP, lexically encodedselectional information interacts with constraints like (17) so as to narrowdown the set of elements that a given lexical head may cooccur with.

The basic structure of declarative clauses is determined by the typedeclarative-head-subject-clause, mentioned earlier, whose relevant super-types are repeated in (19):


(19) phrase

CLAUSALITY

clause

core-cl

decl-cl

HEADEDNESS

hd-ph

hd-subj-ph

decl-hd-su-cl

Kim likes Sandy

Decl-hd-su-cl is a subtype of both decl-cl and hd-subj-ph. Hence claus-es of this type must satisfy the constraints governing both supertypes, aswell:

(20) hd-subj-ph:

hSUBJ h i

i� � , H

�phrase

SUBJ h � i

�

(21) decl-cl:��

HEAD verb

CONT

�proposition

SOA �

�� H

hCONT �

i� � �

Note that (20) specifies local combinatorics (the phrasal head daughtercombines with a subject daughter) and (21) specifies the appropriate seman-tics (a proposition constructed from the state of affairs (soa) provided bythe head daughter.


These more general constraints—applicable to other types of phrase aswell—leave only the following constraint as specific to the type decl-hd-su-cl:

(22) decl-hd-su-cl:

[ ] � � � � H

��

phrase

HEAD

�FORM fin

INV �

��

The feature INVERTED (INV), motivated by lexical considerations discuss-ed briefly in section 7.2 below, is used here to prevent ‘inversion-only’verbs (e.g. first-person aren’t) from appearing in this kind of uninvertedclause.

The type constraints in (20)–(22) combine with each other and withmore general principles outlined above (e.g. the GHFP). In consequenceof these constraints, a VP like (18) will give rise to a declarative clauselike (23) (GHFP effects again illustrated by shading):

(23) S��

decl-hd-su-cl

HEAD �

��

verb

AUX �

INV �

FORM fin

��

SUBJ h i

CONT

�proposition

SOA �

�

��

H

� NP��HEAD

��CASE nom

PER 3rd

NUM sg

��

VP��HEAD �

SUBJ h � i

CONT �

��

Kim likes Sandy


Note that individual constructions do not stipulate the linear order of thedaughters. This is determined by linear precedence rules of greater gener-ality, for example, the following:

(24) LP1: H[word] � X

LP2: � �H

�phrase

SUBJ h � i

�

To complete our sketch of the basic theoretical framework, we mustconsider one more type of phrasal construction—the modifier-head-phrase(mod-hd-ph) sketched in (25):

(25) mod-hd-ph:

[ ] �

HEAD

hMOD �

i, H �

This construction type is both a head-adjunct-phrase and a non-clause.The analysis of modifiers works as follows. Certain words bear nonemp-

ty specifications for the feature MOD (one of the features specified withinHEAD). A modifier-head structure involves the combination of such a mod-ifier with a head daughter whose features are compatible with (or ‘unify’with) the modifier’s MOD value. The adjective never is one such modifier.In virtue of its lexical representation in (26), it may appear as the non-headdaughter in the mod-hd-ph sketched in (27):

(26) never�HEAD

�adv

MOD VP

��


(27) VP��

mod-hd-ph

HEAD �

��verb

AUX �

FORM fin

��

SUBJ A h NP[nom,3rd,sg] i

��

H

�HEAD

�adv

MOD �

�� VP��

fin-ph

HEAD �

SUBJ A

��

never liked Sandy

This VP can be embedded within the decl-hd-su-cl construction to formsentences like Our director never liked Sandy.

The system of grammar we assume here thus generates grammaticalstructures (surface forms) directly. Each piece of a grammatical structureinstantiates some type of construction—just as each node in a well-formedcontext-free tree structure is sanctioned by some grammar rule. The nodesin our structures, however, are labelled by feature structures embodyingsyntactic and semantic information of considerable subtlety. Moreover,though transformations and transformational derivations are entirely elim-inated from the theory of grammar, the organization of phrasal types intoan inheritance hierarchy allows generalizations that cut across families ofconstructions to be succinctly expressed.

3. Not AND Ne-pas AS CONSTITUENT MODIFIERS

When English not negates an embedded constituent, it behaves much likethe negative adverb never. The similarity between not and never is particu-larly clear in nonfinite verbal constructions (participle, infinitival and bareverb phrases), as illustrated in (28) and (29) (cf. Baker 1989):


(28) a. Kim regrets [never [having read the book]].

b. We asked him [never [to try to read the book]].

c. Duty made them [never [miss the weekly meeting]].

(29) a. Kim regrets [not [having read the book]].

b. We asked him [not [to try to read the book]].

c. Duty made them [not [miss the weekly meeting]].

French ne-pas is no different in this regard. Ne-pas and certain other adverbsprecede an infinitival VP:

(30) a. [Ne pas [repeindre sa maison]] est une negligence.

ne not paint one’s house is a negligence

‘Not to paint one’s house is negligent.’

b. [Regulierement [repeindre sa maison]] est une necessite.

regularly to paint one’s house is a necessity

To account for these properties, we regard not and ne-pas not as headsof their own functional projection, but rather as adverbs that modify non-finite VPs. The lexical entries for ne-pas and not include the informationshown in (31).�

(31) not/ne-pas��HEAD

�adv

MOD VP[nonfin]: �

�

CONT NOT ( � )

��

The lexical entry in (31) specifies that not and pas modify a nonfinite VPand that this modified VP serves as the semantic argument of the nega-tion. This simple lexical specification correctly describes the distributional

� Here we assume that both languages distinguish between fin(ite) and nonfin(ite) verbforms, but that certain differences exist regarding lower levels of organization. For exam-ple, prp (present participle) is a subtype of fin in French, whereas it is a subtype of nonfinin English.

For ease of exposition, we will not treat cases where the negation modifies somethingother than VP, e.g. adverbs (not surprisingly), NPs (not many students), or PPs (not in amillion years). Our analysis can accommodate such cases by generalizing the MOD speci-fication in the lexical entry for not.


similarities between English not and French ne-pas: neither element canseparate an infinitival verb from its complements.�� And both ne-pas andnot, like other adverbs of this type, precede the VPs that they modify:

(32) a. [Ne pas V P �inf �[parler francais]] est un grand desavantage

ne not to speak French is a great disadvantage

en ce cas.

in this case

b.*Ne parler pas francais est un grand desavantage en ce cas.

(33) a. [Not [speaking English]] is a disadvantage.

b.*[Speaking not English] is a disadvantage.

c.*Lee likes not Kim.

(34) a. Lee is believed [not V P �inf �[to like Kim]].

b.*Lee is believed to V P �inf �[like not Kim].

Independent principles guarantee that modifiers of this kind precede theelements they modify, thus ensuring the grammaticality of (32)a, (33)a and(34)a, where ne pas and not are used as VP[nonfin] modifiers. (32)b, (33)b,(33)c and (34)b are ungrammatical, since the modifier fails to appear in therequired position—i.e. before all elements of the nonfinite VP.

The lexical properties of not thus ensure that it cannot modify a finiteVP, as shown in (35), but it can modify any nonfinite VP, as is clear fromthe examples in (36):

(35) a.*Pat [not V P �fin�[left]].

b.*Pat certainly [not V P �fin�[talked to me]]

c.*Pat [not V P �fin�[always agreed with me]].

(36) a. I saw Pat acting rude and [not V P �prp�[saying hello]].

b. I asked him to [not V P �bse�[leave the bar]].

c. Their having [not V P �psp�[told the truth]] was upsetting.

�� The exception to this generalization, where pas follows an auxiliary infinitive (n’avoirpas d’argent), is discussed in section 5.3 below.


And much the same is true for French, as the following contrast illustrates:

(37) a.*Robin [(ne) pas V P ��n�[aime Stacey]].

Robin [(ne) not likes Stacey]

b. Il veut [ne pas publier dans ce journal].

‘He wants not to publish in this journal.’

Note that transformational (head-movement) analyses stipulate that nega-tion is generated freely, even in preverbal position in finite clauses, and thata post-negation verb must move leftward because otherwise some ‘need’would be unfulfilled. Proposals along these lines have involved the need tobind a tense variable, the need to overcome some morphological deficien-cy with respect to theta assignment, and so forth. On our account, no suchsemantic or morphosyntactic requirements are stipulated; instead, what isspecified is a lexical selection property. There is no a priori reason, as faras we are aware, to prefer one kind of stipulation over the other. It shouldbe noted, however, that our proposal makes reference only to selectionalproperties that are utilized elsewhere in the grammar.

In addition to these distributional properties, there is further evidencethat not and ne-pas both modify (adjoin to) nonfinite VPs.

3.1. Double Negation

Given the general assumption that modification is recursive, our treatmentpredicts the possibility of double occurrences of negation in infinitivalphrases. This prediction appears to be correct (see Abeille and Godard[henceforth A&G] 1997 and the references cited there), though the accept-ability of some of these examples is somewhat reduced, presumably fornonsyntactic reasons (e.g. the fact that an unnegated phrase conveys essen-tially the same content, but more concisely):

(38) a. Il est recommande de [ne pas [ne pas [travailler]]], malgre le

beau temps.

‘It is advisable to not not work, in spite of the beautiful weather.’

b. Il est recommande de [ne jamais [ne pas [travailler pendant plusde 15 jours]]].‘It’s advisable to never not work more than 15 days.’

The same is true of analogous examples in English:


(39) (Everyone’s turning the offer down, but I’m wavering...)?I can’t believe you would consider [not [not [taking advantage of theoffer]]].

3.2. VP Fronting

In addition, negated VPs can be fronted in French:

(40) a. Le General a ose ne pas obeir aux ordres.

the General has dared ne not to obey to the orders

‘The general dared [not to obey the orders].’

b. [Ne pas [obeir aux ordres]], le General a ose.

(41) a. Je peux ne pas partir a l’ecole immediatement.

‘I can [not go to school immediately].’

b. [Ne pas [partir a l’ecole immediatement]], je le peux.

The preposed sentences in (40)b and (41)b illustrate that ne-pas and theVP it modifies do form a constituent that can undergo VP preposing.

The VP-fronting construction in English is highly stylized and hard todraw conclusions from. Nonetheless, it seems to provide at least a mod-icum of support for the VP-adjunction structure. Negated VPs can some-times front, as in (42):

(42) a. They told him to not divulge the secret and [not divulge the secret]he must, if he ever wants to gain their trust.

b. (?)They suggested that she not go on the mission and [not go onthe mission] she might.

3.3. Clefting

Clefting constructions provide further supporting evidence for VP-adjunction:

(43) a. (?)C’est [ne pas publier dans ce journal] qu’il veut.

‘It is not to publish in this journal that he wants.’

b. Ce que Kelly voudrait, c’est ne pas partir immediatement.

‘What Kelly would like, it’s [not go immediately].’


The example (43)a demonstrates that the negation pas and the VP it modi-fies form a syntactic unit and thus can appear as the focussed element of ace-cleft. Pseudocleft (ce que) constructions display similar behavior, as in(43)b.

The English clefting data in (44) and (45) are harder to interpret, asmany VPs resist clefting entirely, as shown in (46):

(44) a. It’s [not being invited to the party] that they resent.

b. It’s [not to be invited to the party] that they resent.

c. *It’s [not go to the party] that they should.

d. *It’s [not been to the party] that they must have.

(45) a. What they resented was [not being invited to the party].

b. What they resented was [not to be invited to the party].

c. *What they should is [not go to the party].

d. *What they must have is [not been to the party].

(46) a. *It’s [go to the party] that they should.

b. *It’s [been to the party] that they must have.

c. *What they should is [go to the party].

d. *What they must have is [been to the party].

The relevant conclusion about English clefting may be that it is restrictedto those VPs that can also function as NPs (infinitivals and gerunds). Inany case, despite confounding factors, there is evidence in both Frenchand English to support the claim that the negation-nonfinite VP sequencesconsidered here form a syntactic constituent, as we have claimed.

3.4. Scope

It is in general true that a VP modifier is outscoped by a higher verb andis never outscoped by a verb within the VP it modifies. Thus in exampleslike (47) and (48), the finite verb outscopes the adverb:

(47) a. Kim seems [never [to be alone]].

b. Kim seems [not [to like anchovies]].


c. Pat considered [always [doing the homework assignment]].

d. Pat considered [not [doing the homework assignment]].

(48) a. Kim veut ne pas repeindre la maison

Kim wants not to-paint the house

‘Kim wants not to paint the house’

b. Il est recommande de [ne jamais [travailler pendant plus

de 15 jours]].

‘It’s advisable to never work more than 15 days.’

A VP formed via adjunction of not or ne-pas must have negation as itshighest operator (modulo quantification), so that it will be outscoped byany higher predicate that selects it. And in examples like (49), the lowerverb must be outscoped by the negative adverb:

(49) a. [Never [wanting [to speak French]]] is a problem. (�� [Wanting [to[never [speak French]]]] is a problem.)

b. Kim seems [not [to enjoy [reading Proust]]]. (�� Kim seems [toenjoy [not [reading Proust]]].)

The lexical entry for not and ne-pas must therefore include the informationthat the modified element (the content of the element that the negationadjoins to) be within the scope of the negation.

3.5. Coordination

The fact that that not must outscope what it adjoins to also plays a role inexplaining the interaction of coordination and negation:

(50) a. Dana will [[not [walk]] and [talk]].

b. Dana will [not [[walk] and [talk]]].

c. Dana will [[walk] and [not [talk]]].

d. You can [[walk for miles] and [not [see anyone]]].

In each of the examples in (50), the negation modifies a base-form (bse)VP, satisfying the nonfinite specification given above. Because not is a VPmodifier, it may modify either the coordinate VP or one of its conjuncts,thus allowing for the various scopings sketched in (50)a and (50)b. Notice,


by contrast, that if we accept the general assumption that only categorial-ly identical constituents can be coordinated, then these same facts pose adilemma for the NegP hypothesis: the examples in (50) would be coordi-nations of NegP and VP or VP and NegP.��

The nonfinite VP modification of pas exhibits similar behavior:

(51) a. Paul dit ne pas [[lire le journal] ou [regarder la television]].

‘Paul pretends (to) not read the newspaper or watch televison.’

b. Il dit [[ne pas lire le journal] mais [regarder la television]].

‘He pretends (to) not read the newspaper but watch television.’

As represented in the bracketed structure, the VP-modifying negation in(51)a scopes over both conjuncts (see the lexical entry in (31)), whereaspas is outscoped by the conjunction mais in (51)b. As noted by A&G-97,these different scope possibilities follow naturally from the assumptionthat (ne-)pas modifies nonfinite VPs.

4. FINITE NEGATION

The analysis of not and ne-pas as nonfinite VP modifiers provides a straight-forward explanation for much of their distribution, as just illustrated. Wemay simply assume that French and English have essentially the samemodifier-head construction and that not and ne-pas have near-identical lex-ical entries. With respect to negation in finite clauses, however, there areimportant differences between English and French.

4.1. Two Possible Analyses of Finite Negation

It is a general fact of French that pas must follow the finite verb, in whichcase the verb optionally bears negative morphology (ne-marking):

(52) a. Dominique (n’)aime pas Alex.

b.*Dominique pas aime Alex.

�� And even if these nonidentical constituents are somehow allowed, we still lack anexplanation for the impossibility of other cross-categorial coordinations, e.g. CP and IP.One solution to this dilemma might be to posit an additional functional projection such asPolP (Polarity Phrase, Culicover 1991) or � Phrase (cf. Laka 1990). This would of courseentail generating a phonetically unexpressed element as the head of such a phrase in everynonnegative sentence, a consequence that lacks independent justification.


In English, not must follow the finite verb, which must in addition be anauxiliary verb:

(53) a. Dominique does not like Alex.

b.*Dominique not does like Alex.

c.*Dominique likes not Alex.

One approach to the analysis of French finite negation, suggested byDi Sciullo and Williams (1987), involves the claim that not and pas forma morphological unit with a preceding finite verb. Although this accountmight be able to capture the contrast in (52), there is empirical evidenceagainst it. In French, adverbs of doubt or affirmation such as certainement(‘certainly’), apparemment (‘apparently’), peut-etre (‘perhaps’), and thelike, can intervene between the finite verb and the negation pas:

(54) a. Dominique n’aime apparemment pas Ronnie.

Dominique likes apparently not Ronnie

b. Dominique n’a intelligemment pas repondu a la question.

Dominique has intelligently not answered to the question

Moreover, for many speakers, pauses setting off the adverb make sentenceslike (54)b more natural.�� And this casts further doubt on the claim that thefinite verb forms a (morphological) unit with pas.

In English, we find similar problems with the possibility of adverbscoming between the finite verb and not:

(55) a. They will obviously not have time to change.

b. You are usually not thinking about the right problem.

c. They are obviously not good citizens.

Moreover, if not were to form a morphological unit with the precedingfinite auxiliary, then we would incorrectly predict that not would appear ininverted structures along with the verb, as illustrated in (56):��

(56) a. He [would not] leave the city.

�� As pointed out to us by Paul Hirschbuhler (personal communication).�� These inversions, though historically attested and apparently still acceptable in certain

British dialects (cf. Warner 2000), appear to be unacceptable in the American varieties weare familiar with.


b.*[Would not] he leave the city?

c. He [need not] leave the city.

d.*[Need not] he leave the city?

An alternative account of not as a syntactic V-modifier runs into someof the same difficulties, e.g. the problem of why inversions like those in(56)b,d are not possible. Note further that under this modifier analysis, thestructure implied for sentences like (55) would be the nested modificationstructure in (57):

(57) V[fin]

V[fin] Adv[neg]

V[fin] Adv

will obviously not

But making standard assumptions about the relation between modificationand scope (noted in the previous section), this structure gives the wrongsemantics for (55)a, whose scoping is (58)a, not (58)b:��

(58) a. OBVIOUS (NOT (WILL ( ... )))

b. *NOT (OBVIOUS (WILL ( ... )))

In addition, the analysis in (57) provides no mechanism for ruling out mul-tiple finite negation, which seems impossible in both French (see A&G1997) and English:

(59) a. *Paul ne travaille pas pas quand il fait beau

Paul ne works not not when it makes beautiful

‘Paul doesn’t work when the weather’s nice.’

b. *Kim is not not Sandy.

�� Note the scope restrictions on what are normally analyzed as right-adjoined VP-modifiers:

(i) Kim [[[visits Sandy] infrequently] because of the storm].

Here the outermost modifier (BECAUSE) must outscope the lower modifier.


The syntactic V-modifier analysis thus appears deficient in more thanone way. The generalizations about scope of negation seem to be that, first,negation generally outscopes the main verb (though see below) and, sec-ond, an adverb that precedes the negator in a finite clause also outscopesit.

There are two further properties of finite negation to consider before wepresent our analysis. The first concerns constituency; the second has to dowith Verb Phrase Ellipsis (VPE).

4.2. The Constituency of Finite Negation

There is evidence that the modifier-head structure we have posited for ne-pas is not correct for finite negation in French. In particular, constituencytests, such as they are, tell us that in finite negation, pas does not form aconstituent with the following phrase, i.e. that the two kinds of negationshould be distinguished structurally as follows:

(60) a. Modifier pas: b. Finite Negation:

VP[nonfin]

Adv VP[nonfin]

pas

VP[fin]

V[fin] Adv Complement(s)

pas

Several constituent tests, especially VP preposing (discussed also byWilliams (1994)) and clefting in French, support this distinction in struc-ture. Consider the examples given in (61) and (62).��

(61) a. Jean ne veut pas manger d’escargots.

‘John does not want to eat snails.’

b. Manger des escargots, Jean ne le veut pas, (mais ...)

c.*Pas manger des escargots, Jean ne le veut.

(62) a. Je ne peux pas partir a l’ecole immediatement.

‘I cannot go to school immediately.’

b. Partir a l’ecole immediatement, je ne le peux pas.

�� Following Di Sciullo and Williams (1987), Williams (1994) argues, on the basis ofexamples similar to (61), that the finite verb and the negation pas following it forms asyntactic unit. See above for arguments against this view.


c.*Pas partir a l’ecole immediatement, je ne le peux.

The contrast between (62)b and (62)c suggests that pas does not form asyntactic constituent with the VP complement of a finite verb: the twomay not front together.�� This contrasts sharply with the behavior of whatmust be analyzed as the negative modifier, as we saw earlier:

(63) a. Le General a ose ne pas obeir aux ordres.

the General has dared ne not to obey to the orders

‘The general dared not to obey the orders.’

b. Ne pas obeir aux ordres, le General a ose.

c.*Obeir aux ordres, le General a ose ne pas.

Clefting constructions provide further support for this structural dis-tinction. First consider cleft constructions with finite negation:

(64) a. Il ne voudrait pas partir a l’ecole tout de suite.

‘He wouldn’t like to go to school immediately.’

b.*C’est pas partir a l’ecole tout de suite qu’il ne voudrait.

c. C’est partir a l’ecole tout de suite qu’il ne voudrait pas.

Examples in (64) again show that pas does not form a constituent withthe finite verb’s VP complement; hence the two cannot be clefted together.Again, recall the contrasting behavior of the modifier ne-pas discussedabove:

(65) a. Il veut [ne pas publier dans ce journal].

‘He wants not to publish in this journal.’

b.*C’est publier dans ce journal qu’il veut ne pas.

‘*It is to publish in this journal that he wants not.’

c. (?)C’est [ne pas publier dans ce journal] qu’il veut.

‘It is not to publish in this journal that he wants.’

�� One might argue that these facts should be explained on the independent grounds thata phrase with a constituent dependent on ne can never be dislocated. But this argumentis questionable when we consider that ne is usually absent in colloquial French. With thedeletion of ne, (61)c and (62)c are acceptable, but in this case pas clearly functions asconstituent negation.


Finally, pseudocleft (ce que) constructions display a similar contrast betweenfinite negation and modifier ne-pas:

(66) a. Ce que Terry ne voudrait pas, c’est partir immediatement.

‘What Terry wouldn’t like, it is to-go immediately.’

b.*Ce que Terry ne voudrait, c’est pas partir immediatement.

(67) a.*Ce que Kelly voudrait ne pas, c’est partir immediatement.

‘*What Kelly would like not, it is to-go immediately.’

b. Ce que Kelly voudrait, c’est ne pas partir immediatement.

‘What Kelly would like, it is [not to-go immediately].’

Comparable English data unfortunately provide no clear support forthe structural distinction, because, as noted earlier, both VP-Fronting andVP-clefting are highly restricted. However, we may observe that negationalways has narrow scope in fronted examples. In the examples we consid-ered earlier in (42), negation always had narrow scope with respect to thesentence-final auxiliary. But when negation outscopes the finite auxiliary,as in (68)a, then VP-Fronting is never possible; in fact, the negation mustremain unfronted:

(68) They said they wanted to join us,

a. *but not join us they can.

b. but join us they cannot.

The available evidence thus provides a modicum of support for the samestructural distinction in English.

4.3. Negation and VP Ellipsis (VPE)

An important constraint on VPE (first studied in detail by Baker (1971)) isthat it cannot take place immediately after an adverb:

(69) a.*Kim has never studied French, but Lee has always .

b.*Kim has written a novel, but Lee has never .

And, as noted by Kim and Sag (1995) and Warner (2000), constituent nega-tion obeys this same constraint, as expected:


(70) a. *Kim said we should have heard the news, but Lee said that weshould have not . (cf. ... that we should not have )

b. *They haven’t filed their income tax, and to have not meansthey’re in big trouble. (cf. ... and not to have means ....)

c. Kim wants me to go; *Sandy wants me to not . (cf. Sandy wantsme not to .)

One elegant way of accounting for these facts is to treat VPE (follow-ing Sag and Fodor (1994), Kim (1995, 2000), and Sag (ms.,to appear)) assimple complement omission, i.e. as involving no ‘empty categories’. Ifellipsis involves no phonetically unrealized constituents, then the ungram-maticality of these examples is a simple consequence of there being no VPconstituent in (70)a,b for an adverb to adjoin to. Without an element toserve as the syntactic head, there is no way to construct a modifier-headstructure like the one indicated in (71):��

(71) VP

V[aux]

have

*VP

ADVhMOD VP: �

i

always

These predictions of the traceless ellipsis theory are strikingly accurate.�

Although adverbs cannot directly precede an ellided VP, as we havejust seen, in the case of finite negation, not can precede an ellipsis site(Sag 1976, Ernst 1992):

�� Sag and Fodor (1994) and Sag (2000,ms.) reexamine the putative independent empir-ical motivations for phonetically empty categories, including auxiliary contraction, wannacontraction, weak crossover and constraints on the positioning of floated quantifiers.� There is a key contrast between these examples and those where a post-verbal adver-

bial can be stranded:

(i) Kim won’t finish her book on Monday, but she will on Tuesday.

On our approach, following McConnell-Ginet 1982 and the extension in Bouma et al. 2001,the adverbial on Tuesday is selected by the verb essentially like a complement is selected.These examples are thus analogous to pseudogapping of the sort found in (ii):

(i) It doesn’t bother us, but it does them.

Adverbs are discussed in further detail in section 6.


(72) a. Although you want to have another cookie, you may not .

b. Please do that! I will not .

The stranding of not in VPE appears to be possible only when the preced-ing verb is finite and only when not does not express constituent negation.If not were treated as a modifier in (72), then the observed contrast wouldbe quite surprising, as in both cases there would be no VP for not to modi-fy. However, in the flat structure we propose for sentential negation, not isa sister of the finite auxiliary and may simply cooccur with ellipsis of theauxiliary’s complement, as illustrated in (73):�

(73) VPhfin,aux

i

Vhfin,aux

i ADV

could not

Thus all auxiliary verbs in English participate in VP Ellipsis, whichshould be treated as simple omission of the auxiliary verb’s complement.Though constituent negation cannot license ellipsis of the VP, an auxiliarywhose complement has been elided can project a VP that is negated, asshown in (74):

(74) a. Lee may have been studying too much recently, but I think thatKim [may [not [have ]]].

b. Lee may have been studying too much recently, but I think thatKim may have [not [been ]].

These are well-formed because there is a VP that the not modifies (even inthe absence of empty categories).��

� It is interesting that ellipsis-stranded constituent negation (e.g. I should have not )seems to be possible in certain varieties. This may indicate that should have and the likeare being reanalyzed as a single (finite) verb form.�� Following Bresnan (1976), Sag (1976), and Gazdar et al. (1982), we assume that it is

an auxiliary element that ‘licenses’ VP ellipsis, not the head of �P, as suggested by Lopez(1994). Under his assumptions, cases like (74) present a dilemma. If it is assumed that thehead of �P must properly govern the elided (phonetically empty) VP, then the existenceof phonetically overt heads such as have and been would block the negation from head-


4.4. Scope Idiosyncrasy

Finite negation is lexically idiosyncratic in a number of ways. Indeed,much of the literature on English negation, inversion, and ellipsis hasfocussed on the fact that the only elements that allow finite negation arethe modals, the verbs have, be, and do, and a handful of other ‘semiauxil-iaries’, e.g. dare and need. But particular to sentential negation is the fact,documented in detail by Warner (2000) (see also Horn 1972, Gazdar et al.1982, Ernst 1992), that modals exhibit considerable variation with respectto scope of negation.

In the more common pattern—that found, for example, with could, will,deontic may, can, and dare—negation takes wide scope:

(75) a. Paul could not have worked as hard, could he?No, he could not .

b. They will not attend the reception, will they?

c. Kim may not drink the wine on the table. ‘Kim is not permitted todrink.’

Note the polarity of the tags in Warner’s carefully constructed examples.When the polarity of the tags is reversed, we see that these same modalsalso allow not to function as constituent negation, as illustrated in (76):

(76) a. Paul could [not accept the offer], couldn’t he?

b. They will [not accept the offer], won’t they?

c. Kim may [not drink the wine] if she doesn’t like it. ‘Kim is per-mitted not to drink.’

This is of course just what one would expect, if there are two kinds ofnegative construction, as we claim. In addition, one would expect the twoconstructions to cooccur, as they do in the following examples, discussedby Akmajian et al. (1979), Gazdar et al. (1982), and Kim (1995), inter alia:

(77) a. They cannot (just) [not take advantage of that offer], can they?.

governing the VP trace. The only apparent solution, to allow negation to govern the emptycategory across other heads, would violate the head-government condition, presumablystated in terms of Relativized Minimality (Rizzi 1990). It has also been suggested, byZagona (1988), that Infl (Tense), rather than the functional head not, licenses the trace ofthe elided VP. But this again provides no explanation for the contrasts between finite andconstituent negation regarding VPE, nor for the grammaticality of cases like (74).


b. You children may not (simply) [not do your homework] (and stillpass the course).

The idiosyncratic pattern of finite modal-negation scope, where themodal outscopes negation, is illustrated by epistemic may, and deonticmust, should, ought, better, and shall:

(78) a. Kim may not drink the wine if she doesn’t feel like it. ‘Kim pos-sibly won’t drink the wine.’

b. Paul must not accept the offer.

c. They should not have been drinking.

One might think that these examples are instances of constituent negation(which would explain the narrow scope of negation), but they are not. Notethe following instances of VPE:

(79) a. Sandy may not accept the offer and Kim may not either. ‘Kimpossibly won’t accept.’

b. They want Paul to accept the offer, but he must not . ‘He isobligated not to accept the offer.’

c. Should they drink? They should not . ‘They are obligated not todrink.’

As we have seen, VPE cannot strand the not of constituent negation. Hencethese examples must be instances of finite negation, even though the modaloutscopes the negation. Moreover, there is a certain lexical arbitrariness—Warner (2000) shows at length that there is no straightforward way to com-pletely predict these scopal patterns on semantic grounds.

There is parallel idiosyncrasy in French, though there is a smaller classof verbs with meanings like the English modals. In finite negation con-structions, pas usually outscopes the verb, as in (80):

(80) Kim ne peut pas faire ca.

Kim ne-can not to-do that

‘Kim cannot do that.’ (is not able to do that)

But in certain cases, e.g. conditional deontic uses of devoir, we find excep-tions like the English examples just discussed, e.g. the following:


(81) Kim ne devrait pas faire ca.

Kim ne-should not to-do that

‘Kim should not do that.’ (is obliged not to do that)

In the case of the verb falloir and indicative deontic uses of devoir, manyspeakers find the relevant examples to be ambiguous:

(82) a. Kim ne doit pas faire ca.

Kim ne-must not to-do that‘Kim must not do that.’ (is obligated not to do that) or‘Kim is not obligated not to do that’

b. Il ne faut pas crier.

It ne-is-necessary not to-shout‘It is forbidden (necessary not) to shout.’ or‘It is not necessary to shout.’

These peculiarities are thus similar to the English facts, but not identical.We conclude that the scope idiosyncrasy exhibited by French and Englishmust be lexicalized differently in the grammar of each language.�� Thesescope irregularities, we claim, provide evidence for treating both pas andnot in finite clauses as a complement selected by the finite verb,�� a posi-tion corroborated by independent evidence, as we will now show.

5. A LEXICALIST APPROACH TO NEGATION

5.1. Finite Negation

More than one analysis is available that would accommodate the ‘flat’structure we have just motivated for finite negation. Kasper (1994), forexample, proposes a head-complement-adjunct phrase type for analyzing

�� Note further that Modern German exhibits a slightly different pattern. In the followingsentence, due to Stefan Kaufmann (p.c.), negation must outscope mussen, the analogue ofEnglish must:

(i)Kim muß das nicht tun.

Kim must that not to-do.

‘Kim is not obligated to do that.’

�� The argument from scope idiosyncrasy parallels one of Zwicky and Pullum’s (1983)arguments for the lexical treatment of negative contracted forms in English—see below.


the word order freedom of complements and adjuncts in embedded Ger-man clauses. Adapting his proposal to the problem at hand, not or paswould be treated as a modifier, but they would be allowed to appear assister of the head and its complements. An alternative account, followinga long tradition in the HPSG literature,�� involves treating finite negationnot in terms of modification, but rather via lexical selection. That is, finite(auxiliary) verbs would be allowed to select an ADV as a complement (inaddition to their normal complement(s)).

A&G (1997) provide three arguments in favor of the latter proposal forFrench finite negation. First, as noted earlier (see (59) above), only one pasper finite verb is possible. This is behavior characteristic of complements,not modifiers. A&G’s second argument is based on the well known dif-ferences in the realization of pronominal affixes in positive and negativeimperatives:

(83) a. Lis-le!/*Le lis! ‘Read it!’

b. Ne le lis pas!/*Ne lis-le pas! ‘Don’t read it!’

Elements like le in such examples have been shown to be inflectional affix-es, rather than clitics.�� Hence, given the principle of Strong Lexicalismwhich A&G take as axiomatic in the theory of grammar (correctly, in ourview), the two verb forms lis-le and (ne)-le-lis are distinct words of French.If pas were a modifier, it would have to be restricted so as to disallow mod-ification of lis-le. But this restriction, which is quite difficult even to for-mulate, is easily eliminated if (ne)-le-lis and the other negative imperativeverb forms select for an obligatory negative adverbial as their complement.

Finally, A&G note that negative finite verbs differ from their positivecounterparts with regard to the mood of their complement clause:

(84) a. Je crois qu’il est/*soit venu. ‘I think that he has come.’

b. Je ne crois pas qu’il est/soit venu. ‘I don’t think that he has come.’

Again, such contrasts are straightforwardly accounted for if the positiveand negative verb forms are distinct lexical items.

A&G thus propose a lexical rule that gives rise to lexical entries likethe following:

�� See Warner 1993, 2000; Kim and Sag 1995; A&G 1997; and Sag and Wasow 1999.�� See Miller 1991, Auger 1994, and Miller and Sag 1997.


(85) (n’)aime��

HEAD

�verb

FORM fin

�

COMPS h ADVneg , VP[inf] i

SUBJ h NP[nom] i

��

Note that the category ADVneg includes not just pas, but also plus‘(any)more’, point ‘at all’, and guere ‘scarcely’, whose behavior is identi-cal to that of pas in relevant respects. The optional ne-morphology on theverb can be accommodated in much the same way as pronominal affixes(often incorrectly regarded as ‘clitics’) are treated in recent work. (See, forexample, Miller and Sag 1997.)

Once the appropriate finite lexical forms are included in the lexicon (byappeal to the appropriate principles of lexical construction), then the finitenegated structure can be analyzed as a head-complement construction, asillustrated in (86):

(86) VP�HEAD

COMPS h i

�

V��HEAD

hFORM fin

iCOMPS h � , � i

��

�ADVneg �NP

(n’)aime pas Jean

The ADVneg is ordered after the lexical head, just as other complementsare.

Sag (to appear) reaches a similar conclusion about English finite nega-tion, which, he argues, must be understood in the broader context of whathe calls ‘polarized’ verbal forms. That is, English has certain finite auxil-iary verb forms that express either positive or negative polarity. Most famil-iar among these are the not-contracted forms which have lexical entrieslike (87):��

�� This analysis follows in the main Zwicky and Pullum (1983) in assuming that not-contracted finite auxiliaries are formed lexically, not via cliticization. Kim and Sag (1995)propose a lexical rule to achieve this effect; Warner’s (2000) analysis uses lexical types andno lexical rules. Under either analysis, partial formal regularities can be accommodated,


(87) won’t��

HEAD

��verb

FORM fin

POL �

��

COMPS

� VP�FORM bse

CONT �

��

SUBJ h NP[nom] i

��

Here and throughout, the feature POL(ARIZED) distinguishes polarizedauxiliaries from other verbal forms.

A second type of polarized auxiliary selects for a polar adverb, muchlike the lexical entry for French (n)aime just illustrated. For example, theauxiliary can will have a lexical entry like (88):��

(88) can��

HEAD

��

verb

AUX �

POL �

FORM fin

��

COMPS

�ADVpol ,

VPhFORM bse

i�

SUBJ h NP[nom] i

��

exceptional forms (e.g. won’t, don’t) can be listed, anomalies like *amn’t and *mayn’t canbe treated as paradigmatic gaps, and idiosyncratic scope patterns can be accommodated. Inaddition, dialectal variants (e.g. %usen’t, %ain’t) are treated as simple lexical differences.�� The idea of allowing verbs to select for elements that otherwise serve as modifiers

of VPs is closely related to the categorial grammar notion of type raising. In the stan-dard presentation of this idea, an element X that ordinarily serves as a semantic argumentof some modifier Y is assigned a ‘higher order’ meaning F �X� that can take Y as itssemantic argument. Though the function-argument relations are reversed in type raising,the meaning expressed by applying F �X� to Y is exactly the same as the result of applyingY to X . For general discussion of type raising, see Partee and Rooth 1983, Dowty 1988,or Wood 1993. For concrete proposals within HPSG, see Manning et al. 2000 (Japanesecausatives), Bouma and Van Noord 1994 (Dutch complex predicates); Warner 1993 andKim 1995, 2000 (English negation), Miller 1991 and A&G 1997 (French negation), andPrzepiorkowski 1999a (diverse languages). Note that in categorial grammar, type raising isusually presented as a general combinatoric principle. The lexicalist type-raising analysis,by contrast, changes the mode of combination only in limited instances, those controlledby application of the lexical rule. For some discussion of this issue, see Carpenter 1992.


The positive adverbs too and so are also of category ADVpol . Sag’s (toappear) analysis�� thus provides a unified account of the uniform distri-bution of positive and negative polar adverbs, including their inability toiterate or cooccur:�

(89) a. Kim will not believe that.

b. Kim will so/too believe that.

c. *Kim will not not believe that.

d. *Kim will so so believe that.

e. *Kim will not so/too believe that.

f. *Kim will so too believe that.

A further property of ADVpol -selecting auxiliaries is that they cannotbe focused (See Kim 2000.):

(90) a. *They WILL not be there. (Will they?)

b. *Leslie CAN not do that. (Can she?)

c. *Leslie CAN so/too do that.

This ‘antifocus’ property is presumably also lexically registered, either as aconstraint on the lexical type that (88) is assigned to, or else as a conditionon the output of the lexical rule that creates such forms.

Polarized forms like (88) thus project conventional head-complementstructures like (91) by the universal and language-particular principles ofour theory:

�� See also Kim’s (2000) analysis in terms of ‘ADVI ’.� This approach should be contrasted with that of Kayne (2000), who treats emphatic

so, too and not uniformly as functional heads.


(91) VP��fin-vp

HEAD

COMPS h i

��

V��HEAD

��FORM fin

POL �

AUX �

��

COMPS h � , � i

��

�ADVpol �VP

can not do that

Again, the position of the adverbial is determined by general principlesthat order complements after a lexical head.

Note that we treat the ADVpol -selecting verbal forms as positively spec-ified for the feature POL. Since these are derived from basic verbal lex-emes that are [POL �], an ADVpol -selecting form cannot give rise to aform selecting more than one ADVpol . This correctly predicts that Englishallows only one sentential ADVpol per clause. The second not in exampleslike (92) is analyzed as constituent negation:�

(92) Kleptomaniacs cannot NOT steal.

There is semantic motivation for this treatment: if both occurrences of notwere instances of sentential negation, the sentence would mean ‘kleptoma-niacs can steal’. Though (92) clearly has this as an entailment, it is actuallyused to assert something stronger, namely, that kleptomaniacs MUST steal.This is the interpretation predicted by our approach, as the first not negatesthe whole sentence whereas the second one negates only the following VP:steal.

Finally, incorporating Jackendoff’s (1972) observation that focussedauxiliaries idiosyncratically make reference to the ‘affirmation-negation’distinction, Sag proposes an account of the fact that pairs like (93)a,b sharea common reading, one where the disputed proposition that Kim will leaveis being reaffirmed:

� Since the polar adverbs so and too (unlike not) do not function as VP modifyingadverbs, our analysis predicts the nonexistence of sentences like (i):

(i) *Kleptomaniacs cannot so/too steal.


(93) a. Kim will so/too believe that.

b. Kim WILL believe that.

Sag’s analysis posits distinct lexical entries for focussed finite auxiliaries(essentially the positive analogs of not-contracted forms). These entries,again analyzable in terms of lexical types or lexical rules, correlate the aux-iliary focus with the reaffirmation of an appropriate contextually-salientproposition. This analysis of focussed [POL �] auxiliaries plays an impor-tant role in Sag’s treatment of unfocussed do, sketched in Section 7 below.

So far, we have examined the following: (1) evidence for a ‘flat’ struc-ture where not or pas is the sister of the finite verb it cooccurs with; (2)the possibility of stranding not only in finite instances of VP ellipsis; (3)the lexically idiosyncratic nature of the scope of finite negation, (4) theimpossibility of iterating ADVpol in finite constructions; (5) the correla-tion of negative adverbs in French with idiosyncratic positioning of ‘clitic’pronouns (known to be inflectional affixes); (6) selectional differences cor-relating with the presence of finite negation in French; (7) the uniformordering of finite negation in complement position; and (8) a system of‘polarized’ finite auxiliaries in English. All of these phenomena, we havesuggested, are naturally accounted for if finite verbs in both English andFrench are allowed to select these adverbs as complements.��

We have also seen two differences between French and English: (1) allfinite verbs in French allow ADVpol as a complement; in English this phe-nomenon is restricted to indicative auxiliary verbs; and (2) In French, theadverbs in the class ADVpol are all negative (e.g. jamais, plus),�� where-as in English, not and the positive adverbs so, and too all belong to thisclass. A third difference between English and French, which we have notyet motivated, concerns the nature of the constructions in which ADVpol -selecting verbs are realized. We return to this issue in subsequent sections.

�� Note further that adverbial complements are nothing out of the ordinary in many lan-guages, including English. All of the following examples arguably involve adverbials thatare selected as complements:

(i) This book reads *(easily).

(ii) I worded the letter *(carefully).

(iii) The management has treated Sandy *(contemptuously).

�� Hence, we will continue to use the abbreviation ADVneg in discussing Frenchadverbials.


5.2. Have/Be and Avoir/Etre

The position of not and pas is highly flexible in infinitival constructionsheaded by the auxiliaries have/be and avoir/etre. This section extends theproposed analysis to account for the observed patterns.

In English infinitives, the position of not is variable:

(94) a. It was foolish for him not to have been watching more carefully.

b. It was foolish for him to not have been watching more carefully.

c. It was foolish for him to have not been watching more carefully.

d. It was foolish for him to have been not watching more carefully.

Our analysis of not as a modifier of VP[nonfin] is sufficient to account forall the structures in (95).��

(95) a. ... [not [V P �inf � to have been watching more carefully]].

b. ... to [not [V P �bse� have been watching more carefully]].

c. ... to [have [not [V P �psp� been watching more carefully]]].

d. ... to [have [been [not [V P �prp� watching more carefully]]]].

That is, each modified VP in (95) is a nonfinite VP of some sort, and henceeach negated VP in (95) is already predicted to be a possible head-modifierstructure in the analysis we have sketched.��

�� We assume, following Pullum (1982), that the infinitival word to is a defective nonfi-nite ([FORM inf]) auxiliary verb and hence projects a nonfinite VP structure.�� Given the fixed position of functional projections and movement in a head-movement

analysis, it is not easy to capture this ordering flexibility. In an effort to deal with thisproblem, Pollock (1989: 375) suggests that English employs Affix Movement (Chomsky’s(1981) “rule R”) to adjoin the infinitival marker to to the lower VP (in fact onto the initial Vof the VP). (Chomsky (1981: 255–257) applies rule R at S-structure in pro-drop languageslike Italian and Spanish, hence allowing PRO subjects to be ungoverned.) The ordering ofnot to VP in (95)a is derived from the base ordering of to not VP in (95)b via applicationof rule R to the marker to.

There are potential difficulties here posed by the ECP as well as the general problemof motivating such a movement. But even leaving these concerns aside, there remains theproblem of accounting for the position of not in examples like (94)c and (94)d, for neitherrule R nor head movement allows these orderings. Pollock (1989: 375) suggests in a foot-note that the marker to in (94)a should be generated under Tense and the one in (94)b underAgr. But this account also fails to capture the distribution illustrated in (94)c and (94)d.Pollock (1997a) suggests a different solution, with different unresolved consequences.


The possible position of auxiliaries in French infinitive clauses is flex-ible, as (96)a–b and (97)a–b show, a fact that led Pollock (1989) to thepropose the ‘Split Infl’ Hypothesis (Pollock 1989).

(96) a. Ne pas avoir eu d’enfance heureuse est une condition ...

‘To have not had a happy childhood is a condition ...’

b. N’avoir pas eu d’enfance heureuse est une condition ...

(97) a. Ne pas avoir d’idee sur le sujet...

‘Not to have any idea about the matter...’

b. N’avoir pas d’idee sur le sujet...

But our lexicalist account readily accommodates such variation.We begin with the observation that in French the relative position of

infinitival avoir/etre and the negative adverbs pas/jamais, etc. has evolvedover time from Ne V[inf] pas to Ne pas V[inf] (see Hirschbuhler andLabelle 1994). In Modern French, the acceptability of the Ne V[inf] pasordering (e.g. (96)b and (97)b) is restricted to certain conservative (perhapsarchaic) varieties. Thus we might also describe these varieties in terms ofa lexical class of auxiliary verbs whose basic valence pattern includes anoptionial ADVneg, as shown in (98):

(98) etre/avoir,...��HEAD verb

SUBJ h NP i

COMPS h (ADVneg ,) VP i

��

This proposal, similar to the one made by A&G (1996), correctly allowsstructures like (99)b:

(99) a. V P [Ne pas V P �inf �[avoir d’idee sur le sujet]] ....

b. V P �inf �[V �inf �[N’avoir] Adv[pas] NP [d’idee sur le sujet]] ...

The auxiliary/nonauxiliary distinction has eroded in most varieties of Mod-ern French, which do not allow such examples because no lexical classallows the basic valence pattern shown in (98).

This explanation also applies to the distribution of pas in modal con-structions. Hirschbuhler and Labelle (1994) note that, as in the case ofinfinitive auxiliaries, the ordering Modal-V[inf] + pas, as in (100)b, hasgradually given way to pas + Modal-V[inf], as in (100)a:


(100) a. Je pensais ne pas devoir partir a l’ecole aujourd’hui.

I thought ne not ‘have to’ go to school today

‘I thought I must not go to school today.’

b. Je pensais ne devoir pas partir a l’ecole aujourd’hui.

The present analysis allows us to treat modal negation like (100)b simplyby assuming that devoir, vouloir and the like are auxiliary verbs in thevarieties that allow such sentences.��

Since our analysis in no way relies on role assignment (see below), itavoids a further unintuitive consequence of Pollock’s (1989) head-movementanalysis. As Hirschbuhler and Labelle (1994) point out, Pollock’s analy-sis entails that even the main verb usage of verbs like vouloir (‘want’)failed to theta-assign in older varieties of French. We treat the variationand diachrony very simply here—in terms of erosion of the auxiliary/nonauxiliary distinction. Changes in certain syntactic phenomena such asclitic climbing and ‘Aux-to-Comp movement’ imply that modals havebecome recategorized as lexical (rather than auxiliary) verbs.�� Whenmodals are treated as auxiliaries, they may take an ADVneg as their com-plement, predicting that pas can follow the infinitival modal, as in (100)b.

5.3. Negation with Semi-Auxiliary Verbs

The English negative element not can also follow the so-called semi-auxiliary verbs dare and need. The contrast between need and dare giv-en in (101) and (102) shows that dare assigns a theta role (AGENT, let usassume) to its subject:

�� Pollock, by contrast, suggests (1989: 390) that these verbal elements (which have fullverbal paradigms) must be treated essentially as modifiers.�� Such phenomena, though attested in earlier stages of French as in (i), are not available

in Modern French as shown in (ii) (data from Roberts (1994: 233)):

(i) a. Nous lui devons rendre gloire

We to-him must give glory

b. Ayant ce bon homme fait tout son possible...

Having this good man done everything possible...

(ii) a.*Je le peux faire

I it can do

b.*Ayant Jean fait cela,...

Having John done that...


(101) a. It need not be raining.

b. It need not bother you that we will be absent.

c. There need not be any independent motivation.

(102) a.*It dare not be raining.

b.*It dare not bother you that we will be absent.

c.*There dare not be any independent motivation.

The possibility of its theta-role assignment is further attested by exampleslike (103):

(103) a. The child dare not contact her.

b. They dare not move.

In terms of theta-roles, semi-auxiliary verbs are thus different from (most)modals or have and be, implying that the negative placement in this con-struction cannot be attributed to theta theory, as suggested by Pollock(1989).��

The solution made available within our analysis turns on the featureAUX, which semi-auxiliary verbs such as dare and need may be positivelyspecified for. This predicts that they allow not as a complement.�� Thusnot in (102) and (103) can be either a nonfinite VP modifier or a comple-ment. As a complement of dare, not is predicted to allow VP ellipsis. Thisprediction is correct:

�� If the possibility of a verb’s movement depended on whether or not it can assign atheta role, then verbs like dare would require an explanation. One approach might be togenerate dare directly under Tense, like other non-theta-assigning modals, but this raisesnew problems, e.g. how theta roles can be assigned to arguments embedded within VP.Though technical solutions to this problem may exist, examples like (i) cast further doubton the attempt to explain alternations of verb position in terms of theta theory:

(i) They dared not carry out their threat. (from Quirk et al. 1985: 138)

The problem here is that dare occurs in inflected form, yet assigns a theta role to the phrasecarry out their threat. The inflection on dare would have to be sufficiently ‘strong’ totransmit theta role assignment, contradicting Pollock’s claim that this can never happen inEnglish. For further arguments that morphological strength in Pollock’s sense is historical-ly at odds with observed morphological richness, see Baker 1991.�� In many varieties, the [AUX �] dare (taking an infinitival VP complement) is in com-

petition with the [AUX �] dare.


(104) a. Lee wants me to contact them, but I dare not .

b. It was suggested that we contact them, but we dared not .

6. ADVERBIAL POSITIONS

6.1. English Adverbial Positions

The modifier-head construction formulated in (25), repeated here as (105),allows many kinds of modifier-head phrases, depending on what constraintsare included in the lexical entries of English modifiers:

(105) mod-hd-ph:

[ ] �

HEAD

hMOD �

i, H �

In finite clauses, then, our grammar already allows all the following possi-bilities:

(106) ShFORM fin

i

ADV� ShFORM fin

i

NP VPhFORM fin

i

ADV� VPhFORM fin

i

VhFORM fin

i (not) VPhFORM nonfin

i

ADV� VPhFORM nonfin

i


All of these positions are attested (though it is sometimes unclear whethera particular adverbial should be treated as in a given position or as aparenthetical).�

In the absence of linear precedence constraints ordering modifiers andhead daughters, the account presented so far might also allow adverbs tofollow the phrases they modify, as shown in (107):

� Cinque (1999) presents a competing conception of clause structure. His head-movement analysis assumes crucially that adverbs are base-generated in different func-tional positions (according to their meaning) and can be moved to other non-wh-operatorpositions. These assumptions lead Cinque to postulate more than two dozen different func-tional projections (see Cinque 1999: 106). On his theory, the hierarchies of adverbial spec-ifiers and clausal functional heads also match in a one-to-one fashion. The surface order ofadverbs thus depends on the the order of functional heads in UG, features linking a specificclass of adverbs to each head, and the existence of movement operations.

A system like Cinque’s could solve some of the problems that we discuss for Pollock’s(1989) analysis. However, as Cinque (1999: 127) himself admits, such a system raisesnumerous questions. For example: should the entire array of functional projections bepresent in every sentence of every language, even when there is no overt morphologicalrealization? Is the order of functional projections really invariant cross-linguistically, andif not, how much variation should UG allow? In addition to these concerns, there is thefurther worry that Cinque’s system must appeal to a large number of abstract, phantomformatives, thus making the task of language learning more complicated than it would beunder a system like ours. The proposals we make here lead to a grammar that has at leastcomparable empirical coverage, but which is constraint-based, surface-oriented, and hencemore likely to be learnable on the basis of experience—a highly desirable result. For adiscussion of further problems facing Cinque’s analysis, see Ernst 1998.


(107) ShFORM fin

i

ShFORM fin

i

NP VPhFORM fin

i

VPhVFORM fin

i

VhFORM fin

i (not) VPhFORM nonfin

i

VPhFORM nonfin

i ADV�

ADV�

ADV�

There are clearly modifiers that appear in the ADV� position – typical-ly preceded by a pause. But since these have very little interaction withthe issues of the auxiliary system, we will pay little attention to themhere. More relevant are the ADV� and ADV� positions. In a number ofrecent studies,� it has been argued that the reanalysis of adverbials asoptional complements suggested by McConnell-Ginet (1982) for manneradverbials should be extended to certain other kinds of adverbials, includ-ing temporals and locatives. The evidence for this ‘adverbs as comple-ments’ analysis is diverse and intriguing, ranging from the case-markingpatterns in languages such as Korean and Finnish (Wechsler and Lee 1996,Przepiorkowski 1999a) to the morphological marking of long-distancedependencies in Palauan, Chamorro, and Irish. Bouma et al. (2001) arguethat a unified analysis of English extraction constructions (one that elim-inates separate stipulations for subject, complement and adjunct extrac-tion) is possible precisely if one treats postverbal adverbials as dependents

� See Manning et al. 2000; Bouma and Van Noord 1994; Warner 1993, 2000, Kim1995, and Bouma et al. 2001, Miller 1991 and A&G 1997; Wechsler and Lee 1996;Przepiorkowski 1999a,b.


selected by the verb, arguably eliminating the need for true modifiers tooccur in positions ADV� and ADV�.��

Thus assuming that at least locative, temporal, and manner adverbialsare selected as optional complements, and hence may be extracted by thesame mechanisms that account for complement extraction, we get struc-tures like (108)a and corresponding structures like (108)b, where the com-plement appears in the ADV� position as an extracted element (a ‘filler’),rather than as a modifier:

(108) a. ShFORM fin

i

NP

the doctor

VPhFORM fin

i

VhFORM fin

i

examined

NP

Sandy

ADVhLOC,TMP,MNR,...

i

��

in Rome

todayreluctantly

...

��

�� Note that the syntactic distinction between complements and modifiers has no exactcorrespondence to any semantic distinction. In fact, given the semantics of type-raisingdiscussed in footnote 27 above, an adverbial selected as a complement by a verb may yieldexactly the same semantics that would result if that adverbial were modified the verb.


b. ShFORM fin

i

� ADVhLOC,TMP,MNR,...

i

��

in Rome

todayreluctantly...

��

S�FORM fin

SLASH f � g

�

NP

the doctor

VP�FORM fin

SLASH f � g

�

V�FORM fin

SLASH f � g

�

examined

NP

Sandy

This basic approach deals with a wide range of adverbials that can occurin ADV� position as well as postverbally.�� Nonetheless, there are variouskinds of adverbials (e.g. unfortunately, obviously, usually, apparently,...)that are not allowed as optional complements, but which may appear asmodifiers in the ADV�, ADV�, or ADV� positions:��

�� A residual concern, pointed out by Bob Levine (in remarks made at the 2000 BerkeleyFormal Grammar Conference), involves sentences like the following:

(i) Leslie entered the room, grabbed a book, and jumped into bed in 30 seconds flat.

(ii) Terry sang bizarre sea chanteys and danced in circles for two hours.

In these examples, the adverbial can outscope the conjunction, a fact that follows nat-urally from the standard modificational treatment of these adverbials, but which seemsinconsistent with the ‘adverbs as complements’ analysis. It should be noted, however, thatanalogous examples can be constructed involving uncontroversial complements:

(iii) Leslie sent flowers and subsequently brought candy to thirty hospital patients.

(iv) Terry faxed her mother and (then) mailed my sister thirty tasteless jokes.

Though the proper treatment of all these examples remains somewhat mysterious, it is notapparent that examples like (i) and (ii) are inconsistent with the analysis of the adverbs inquestion as complements.�� This should be compared with Cinque’s (1999) analysis, where ADV� adverbs like

(un)fortunately, regrettably, surprisingly, and (un)expectedly are generated in the specifier


(109) a. Kim will not run *(,) usually.

b. Usually Kim will not run.

c. Kim usually will not run.

d. Kim finds them to usually be trustworthy.

This pattern follows if (1) the MOD value of these adverbials underspec-ifies the SUBJ value (i.e. is compatible with [SUBJ hNPi] or [SUBJ h i] and(2) we have a linear precedence rule like (110):

(110)LP3:

hMOD �

i�

� YPhSUBJ h XP i

i

In this treatment, VP-adjoined adverbials must precede the phrase theymodify, but an S-adjoined adverbial may follow the sentential head. Theintent is to allow (109)a as an S-modified structure, assuming the requiredpause is characteristic of right-adjoined S-modifiers.

There are also adverbials (such as never, merely, really, almost, etc.)that only appear in the ADV� or ADV� position:��

(111) a. *Kim opened the door merely.

b. *Merely Kim opened the door.

c. Kim merely opened the door.

d. We want him to merely open the door.

Again, these are not selected as optional complements of the verb andhence cannot undergo extraction. Rather, these elements appear as mod-ifiers in only the appropriate places because they are lexically specified asin (112):

(112)��MOD

VPhSUBJ h XP i

i��

position of an evaluative mood head. Adverbs such as usually and regularly are in thespecifier position of the habitual aspect head.�� Cinque’s (1999) analysis generates adverbs such as almost and merely in the specifier

position of prospective aspect. Movement processes are required to generate two differentpositions for these adverbs with the same interpretation.


Each class of adverbials instantiates a lexical type where a constraint onthe MOD value may be stated. Our approach thus predicts there will begeneralizations over lexical classes, but it allows syntactic or semanticexceptionality.��

This observation provides a new perspective on issues discussed inthe literature. For example, Pollock (1989) seeks to explain contrasts like(113)a,b in terms of head movement.��

(113) a.*John completely will lose his mind.

b.*John completely has lost his mind.

Pollock’s (1989) explanation for the contrasts in (113) involves the claimthat completely can be adjoined only to VP, not to TP. Because the aux-iliaries move to Tense in his system, VP-adjoined adverbs like complete-ly cannot precede them. But this cannot be the right solution, as pointedout by Iatridou (1990), because of the positional constraints on the adverbcompletely in infinitive clauses:

(114) a. Mary is believed to be completely revising her dissertation.

b.*Mary is believed to completely be revising her dissertation.

According to the head-movement analysis, (114)b would be the unmovedversion of (114)a. In (114)a, be moves to Agr across the adverb completely,as shown in the derivation of (115). Since verb movement is optional ininfinitive clauses (see (116) and (117)), the verb be does not have to move,and may stay in situ. But this then incorrectly predicts that example (114)bshould be acceptable.

�� For example, the adverb completely imposes semantic conditions on the phrase that itmodifies, as the following examples suggest:

(i) a.*Leslie completely remained at home.

b. Leslie remained completely silent.

c. That suggestion seemed to be completely off the wall.

Adverbs like completely can only modify phrases whose semantics is in a certain sense‘gradable’.�� According to Cinque (1999: 100–104), completamente in Italian and completely in

English both occupy two distinct functional specifier positions, associated with two distinctinterpretations. This analysis may solve certain of the problems encountered by Pollock’s(1989) analysis.


(115) TP

T AgrP

to Agr VP

Adv VP

completely V ...

be

In order to exclude cases like (114)b, the movement of be to Agr mustsomehow be made obligatory, which raises the crucial question of howmovement can be made sensitive to particular lexical items.�� And if themovement of be or have to Agr were obligatory in all cases in English, wewould not be able to generate sentences like (116)a and (117)a, where theadverb precedes the auxiliary verbs:

(116) a. Mary is believed to frequently have criticized Bill.

b. Mary is believed to have frequently criticized Bill.

(117) a. Mary is believed to frequently be criticizing Bill.

b. Mary is believed to be frequently criticizing Bill.

It might be suggested that adverbs preceding the infinitive have and be arebase-generated as adjoined to TP, whereas those following them are base-generated adjoined to VP. But this solution would undermine the very exis-tence of verb movement, since the relevant surface orderings could then beobtained without any movement at all, as argued by Iatridou (1990: 555).

Our treatment encounters no such difficulties. Because the surface struc-tures are generated directly, different adverbs are ‘base’-generated in dif-ferent surface positions according to the syntactic and semantic specifica-tions in their MOD value. This method of analysis, endorsed in essenceby Iatridou (1990), Bouchard (1997), and A&G (1996) (among others),provides a simple way of capturing the distribution of adverbs. The adverbfrequently lacks the semantic restriction of completely and hence may mod-ify any of the VPs that follow it in (116)a,b and (117)a,b. The verb be—whether finite or nonfinite—heads a VP that is semantically inconsistent

�� It should be noted that Pollock (1997a) attempts to answer some of Iatridou’s criti-cisms by appeal to a version of ‘checking theory’.


with the semantic demands of completely. Hence Pollock’s contrast in(113) is related to and (semantically) accounted for in just the same wayas the contrast in (118):

(118) a. Mary is believed to be [completely [revising her dissertation]].

b.*Mary is believed to [completely [be revising her dissertation]].

Our grammar relies on lexical information and a carefully articulatedtheory of lexical organization. Once the lexical information is properlyorganized, the appropriate surface phrases are directly generated by thetheory of phrasal constructions (the ‘barless’X-theory) in accordance withuniversal principles such as the GHFP and valence theory.

6.2. French Adverbial Positions

The basic structure of the French clause is much the same as that of English.A&G-97 posit head-subject and modifier-head constructions that are essen-tially the same as those discussed above for English. Many adverbials inFrench are thus lexically specified in such a way (via their MOD value)that they can modify only phrases satisfying certain conditions, thus allow-ing the problem of restricted distributions to be approached lexically, as inEnglish.

In the head-movement analysis, all French verbs undergo movement toT, whereas in English, only the auxiliary verbs have and be undergo thisprocess. The head-movement analysis thus predicts that the English/Frenchauxiliaries be/etre and have/avoir will show similar behavior. But there aredifferences between English and French, especially with respect to adverbposition. As Pollock (1989: 370) notes, English adverbs such as rarely,often, and seldom, can precede the auxiliary verbs have or be:

(119) a. My friends rarely/often/seldom are unhappy for long periods.

b. My friends rarely/often/seldom have helped me.

However, the corresponding French adverbs souvent/rarement cannot appearbefore etre and avoir:

(120) a.*Mes amis souvent/rarement sont malheureux tres longtemps.

My friends often/rarely are unhappy for long periods

b. Mes amis sont souvent/rarement malheureux tres longtemps.


(121) a.*Mes amis souvent/rarement m’ont aide.

My friends often/rarely have helped me

b. Mes amis m’ont souvent/rarement aide.

To address this contrast, Pollock (1989: 370) suggests another parame-ter distinguishing the two languages: English has two adverb positions—the VP-adjoined and TP-adjoined positions, whereas French has only VP-adjoined adverbs.��

Under our approach, however, it is a general property of many Frenchadverbs that they modify only nonfinite VPs. We thus unify the account of(120) and (121) with that of examples like the following:

(122) a.*Mes amis [souvent/rarement [V P �fin� se voient.]].

b. My friends [rarely/often/seldom [V P �fin� see each other]].

Our analysis also avoids a problem for Pollock’s (1989) analysis thatwas pointed out by Kayne (1991). This concerns examples like thefollowing:�

(123) a. John is never satisfied.

b. John never is satisfied.

(124) a. Jean (n’)est jamais content.

b. *Jean jamais (n’)est content.

Given an analysis like that of Pollock (1989), in (123)a and (124)a theauxiliaries move to the topmost functional position. This does not predictthe contrast between (123)b and (124)b. Main verbs behave differently:

(125) a. John never seems happy.

b. *Jean jamais (ne) semble heureux.

�� In allowing only the VP-adjoined position for French adverbs, this system fails tocapture the possibility of ordering adverbs between a finite verb and pas. See section 4.1for examples.� See also Pollock 1997a.


As Kayne notes, we presumably want to have a uniform account of thecontrast between (123)b and (124)b and the contrast between (125)a and(125)b. Yet such an account is unavailable in Pollock’s system.

In our theory, where the lexical properties of never determine its posi-tion, it is straightforward to provide a uniform account of its behavior:never may modify any VP—finite or not. The adverb jamais, by contrast,behaves like pas, as in our earlier discussion. That is, jamais can be thenegative complement of a finite verb, as shown in (124a); and ne-jamaiscan modify a nonfinite VP, as shown in (126):

(126) Jean regrette de [ne jamais [venir a Paris]].

Jean regrets of ne-never to-come to Paris

‘Jean regrets never coming to Paris.’

A&G-97 consider a number of different adverb classes. They argue thatcertain adverbs (which they call simply ‘V-adverbs’) have a dual function,as negative adverbs do. V-adverbs like trop ‘too much’, peu ‘little, assez‘enough’, and a peine ‘hardly’ may adjoin to the left of an infinitival verb(i.e. to a word), as in (127)a. Alternatively, they may be selected by a verbas an optional complement, as in (127)b:

(127) a. Paul croit [[bien [declamer]] Proust].

Paul believes well of-to-recite Proust

‘Paul believes he recites Proust well.’

b. Paul croit [declamer bien Proust].

Paul believes of-to-recite well Proust

‘Paul believes he recites Proust well.’

As A&G show, this dual analysis is motivated by a number of otherwisepeculiar facts, for example the contrast between (128)a,b:

(128) a. Paul esperait mieux manger et boire.

Paul hoped better to-eat and to-drink

‘Paul hoped to eat better and to drink.’

b. Paul mange et boit mieux aujourd’hui.

Paul eats and drinks better today

‘Paul eats and drinks better today.’

As the gloss in (128)a indicates, preverbal mieux cannot outscope theconjunction—it must be lexically attached. However, the example in (128)b,


like its English counterpart, is ambiguous. This ambiguity follows natural-ly if one assumes that postverbal occurrences of mieux and the like arecomplements selected by the verbal head ((128)b may then involve eitherV coordination or VP coordination).

A&G-97 do not spell out their full account of French adverbials, butit is reasonable to assume, following Bouma et al. (2001), that extractableadjuncts (temporal, locatives and some manner adverbials, as well as thepostverbal V-adverbs) should also be treated as complements selected bythe verb. Assuming these adverbials are selected as complements, it is nosurprise that they can sometimes be ordered between the verb and its othercomplements, as in examples like (129):

(129) a. Paul ecoutera attentivement l’orateur.

Paul will-listen(-to) attentively the-speaker

‘Paul will listen to the speaker attentively.’

b. Marie lit souvent Proust.

Marie reads often Proust

‘Marie often reads Proust.’

Indeed, the ‘adverbs as complements’ style of analysis—defended at lengthby Przepiorkowski (1999a,b)—is able to predict many facts about linearorder from independently motivated ordering principles, e.g. the LP rulesconsidered earlier.

But there are certain other adverbials that can separate verbs from theircomplements, e.g. evidemment ‘obviously’, probablement ‘probably’,sincerement ‘sincerely’, malheureusement ‘unfortunately’:

(130) a. Paul a apparemment donne ce livre a Marie.

‘Paul has apparently given this book to Mary’

b. Paul a donne apparemment ce livre a Marie.

c. Paul a donne ce livre apparemment a Marie.

d. Paul a donne ce livre a Marie apparemment.

(131) Dominique aime evidemment Ronnie.

Dominique likes evidemment Ronnie

‘Dominique obviously likes Ronnie.’


These adverbs are not treated as complements by A&G-97. Their reluc-tance to treat them as such is based (Daniele Godard, personal communi-cation) on the fact that these elements do not extract. Hence they do notfit the general pattern observed by Bouma et al.—that the only extractableadverbials are those that can also be selected as complements.

To deal with examples like (130) and (131), A&G-97 adopt Kasper’sproposal to replace the head-complement construction with a head-complement-adjunct construction. This phrasal type allows adverbials to appear as sis-ters of the head and its complements, but does not require that the headselect those adverbials as complements. LP rules, partly stated in terms ofthe key notion of lite vs. non-lite elements developed in A&G 2000, allowlite adverbials to precede their complement sisters. In A&G’s analysis, theVP in an example like (131) is assigned the following structure:

(132) VP�HEAD

COMPS h i

�

V�HEAD

hFORM fin

iCOMPS h � i

� Adv �NP

aime evidemment Ronnie

But in fact, it is possible to provide a uniform account of the dual posi-tioning of French adverbials in terms of the adjunct-complement distinc-tion. In short, adverbs have various MOD values, which means that whenthey function as modifiers, they adjoin to diverse kinds of verbal con-stituents: V, VP, S, or some combination of these. In addition, subjectto various lexically modulated conditions, adverbs may appear as comple-ments selected by the verb. In this case, a flat head-complement structurelike (132) results. The various constraints on adverb order are then handledvia LP rules (which only affect the relative order of sister constituents). Thefact that certain complement adverbials cannot be extracted is accountedfor by placing restrictions on the type adv-gap, which is the type that averb’s adverbial argument must assume when it corresponds to an extract-ed element (assuming the extraction analysis of Bouma et al. 2001). Webelieve that the system just sketched simplifies the account of A&G-97,while preserving its insights.

There are interesting facts about the distribution of French adverbs andnegation that support an approach of this kind. Pas behaves differently


from adverbs like a peine in several respects, in particular in infinitiveclauses:

(133) a. Parler a peine le francais est un grand desavantage en ce cas.

to speak hardly French is a great disadvantage in this case

b. A peine parler le francais est un grand desavantage en ce cas.

(134) a. Ne pas parler le francais est un grand desavantage en ce cas.

ne not to speak French is a great disadvantage in this case

b.*Ne parler pas le francais est un grand desavantage en ce cas.

The distributional contrast between these two types of adverb is the basicmotivation for Pollock’s proposal to split Infl into Tense and Agr and hisassumption of ‘short movement’ in infinitives, as sketched in (135):

(135) TP

T[�finite] NegP

pas AgrP

Agr VP

Adv VP

a peine V ...

parler

The short movement from V to Agr is meant to apply to auxiliaries andto lexical verbs alike. But given the assumption that [�finite] is opaqueto theta role assignment, theta-assigning verbs cannot move from Agr to[�finite] Tense, since this would yield violations of the theta-criterion. Inthis way, the head-movement analysis bars the main verb from precedingthe negation pas in infinitive clauses, e.g. in (134)b. But the movement toAgr is permitted, allowing the verb to precede adverbs like a peine.

An unfortunate consequence of Pollock’s system is that it is not ableto describe dual adverb positions in a clause, as pointed out by Iatridou(1990). There is no other position between TP and VP where anotheradverb can be located in examples like (136):


(136) a. Frequemment parler bien le francais est un grand avantage.

frequently speak well French is a great advantage

‘To frequently speak French well is a great advantage.’

b. Frequemment parler parfaitement le francais...

frequently speak perfectly French ...

‘To frequently speak French perfectly,...’

And the ordering of manner adverbs shown in (137) illustrates a seconddefect of the head-movement analysis of French adverbs, as pointed outby A&G:

(137) a. ?*Attentivement lire ce texte est une condition pour reussir

l’examen.

‘Carefully to read the text is a condition for succeeding the

exam.’

b. Lire attentivement ce texte est une condition pour reussirl’examen.

c. Lire ce texte attentivement est une condition pour reussirl’examen.

d.*Attentivement avoir lu ce texte est une condition pour reussirl’examen.

e. Avoir attentivement lu ce texte est une condition pour reussirl’examen.

f. Avoir lu attentivement ce texte est une condition pour reussirl’examen.

g. Avoir lu ce texte attentivement est une condition pour reussirl’examen.

Unlike sentential adverbs, manner adverbs like attentivement cannot pre-cede an infinitive (lexical or auxiliary) verb, as shown in (137)a (seeBouchard 1997). In order to block such examples, the head-movementanalysis must find some way to make verb movement to Agr obligatoryin just this environment. But it remains obscure how Move-� (the central


explanatory device assumed in the theory) can be made obligatory in suchparticular environments.�

A related issue concerns the scope of manner adverbs, as noted by A&G(1994a, b). Consider (138):

(138) Jean a attentivement ecoute son professeur et pris des notes.

‘John has attentively listened to his teacher and taken notes.’

Assuming that the adverb adjoins to a VP, we would expect the man-ner adverb attentivement to be able to scope over the whole coordination.However, in such examples, the adverb’s scope is confined to the first con-junct; thus (138) conveys only the reading that can be paraphrased as (139):

(139) What John did was listen to his teacher attentively and take notes.

Moreover, the existence of an ordering restriction between certainadverbs casts further doubt on the existence of a VP-adjoined analysis,as noted by A&G (1994a,b):

(140) a. ??Jean a bruyamment immediatement contre-attaque.

‘Jean has loudly immediately counter-attacked.’

b. Jean a immediatement bruyamment contre-attaque.

In an analysis like Pollock’s, where all adverbs are taken to be VP-adjoined,it is unclear how to state the condition that a time adverb must precede amanner adverb.

Given these observations about the position, scope, and word orderproperties of French adverbs, it seems necessary to treat postverbal adverbsin general as sisters of the verbal head. Under our proposal, principles oflinear precedence have access to the lexical head, its complements, and theVP-modifiers, as these are all sister constituents.

We have already outlined our account of negative adverbs. These areinherently modifiers of nonfinite VPs that are also pressed into service asthe optional complement of a finite verb. As for adverbs likea peine (againfollowing A&G), these may be treated uniformly as V modifiers, specifiedlexically as in (141):

� Another alternative might be to generate such manner adverbs in VP-final position,and to assume rightward movement of the object NP, as Pollock (1989: 379–381) suggests.But this again makes the grammar more complicated by allowing an additional movementprocess whose effect is to produce spurious ambiguities (given verb movement) in manycrucial cases (see A&G 1994a).


(141) ADV�MOD V: �

CONT ADV’ ( � )

�

This proposal makes available an account of all cases of short movementin terms of structures like (142):

(142) V

V[inf ] ADV

voir a peine

It also immediately provides a treatment of contrasts like the following:

(143) a. Ils ont a peine donne du secour aux enfants.

They have barely given help to the children

b. Ils ont donne a peine du secour aux enfants.

They have given hardly help to the children

c.*Ils ont donne du secour a peine aux enfants.

They have given help hardly to the children

(143)c is ungrammatical precisely because a peine is not adjacent to, andhence not adjoined to, any verb. Notice that all of this follows from theinteraction of the various principles of HPSG and the partial lexical infor-mation sketched in (141). No other machinery is required.

As for adverbs of time and manner, once these are treated as sistersof the lexical head and its complements, we will have structures like thefollowing:

(144) VP

V ADV[MNR] NP NP

donner gentiment ce livre a Marie

(145) VP

V NP ADV[MNR] NP

donner ce livre gentiment a Marie


The advantage of such flat structures is that they provide an immediatesolution to the problems that face any account stated in terms of uniform-ly adjoined or hierarchical structure. LP rules governing relative order ofadverbs can be stated simply, e.g. as in (146):

(146) ADVneg � ADV[...]

The LP rule in (146) is intended to rule out unwanted examples like (147):

(147)*Dominique (n’)ecrivait hier pas de lettres.

Dominique wrote yesterday not letters

But (146) does nothing to block the rather free distribution of many typesof adverbs within the VP, as illustrated in (148) and (149).��

(148) a. Paul a gentiment donne ce livre a Marie.

‘Paul has gently given this book to Mary’

b. Paul a donne gentiment ce livre a Marie.

c. Paul a donne ce livre gentiment a Marie.

d. Paul a donne ce livre a Marie gentiment.

(149) a. Paul a apparemment donne ce livre a Marie.

‘Paul has apparently given this book to Mary’

b. Paul a donne apparemment ce livre a Marie.

c. Paul a donne ce livre apparemment a Marie.

d. Paul a donne ce livre a Marie apparemment.

Again the key to the analysis is the sisterhood of the relevant adverbs,lexical heads and complements.

There is considerably more to be said about the grammar of Frenchadverbials. Nonetheless, we hope that this brief exposition is sufficient tomake clear how the lexicalist analysis is able to account for a wide rangeof data about French adverb position—data that appear to defy descriptionunder the assumptions of the head-movement approach.

�� We assume here, following A&G 1994a, 1997, and 1998 that the auxiliary verbs avoirand etre serve as lexical heads of flat VPs, i.e. they take a lexical participle and the partici-ple’s complements as its own complements.


7. FURTHER ISSUES

Thus far, we have sketched and justified our treatment of constituent andfinite negation in French and English. Moreover, we have embedded thisaccount within a more general analysis of adverb position in both lan-guages. In this section, we consider two further theoretically critical mat-ters that interact with the grammar of negation: the auxiliary do and inver-sion constructions.

7.1. ‘Dummy’ Do

The English auxiliary do is commonly thought to require transformationalanalysis of some kind (or else OT-style ‘optimization’) in order to expressthe traditional wisdom (see, for example, Grimshaw 1997) that “do is pos-sible only when it is necessary.” Some such generalization has long beenthought to be true of English—that do appears only in contexts wheremovement of tense onto a following verb is prevented in negation, inver-sion, contraction, or ellipsis constructions (the NICE constructions—cf.Quirk et al. 1985 and Warner 1993). Following Sag (to appear), we providea lexicalist analysis of do that employs no techniques beyond the simulta-neous satisfaction of grammatical constraints.

In previous lexicalist treatments of the English auxiliary system (fromHudson 1976 to Kim and Sag 1995), it has been assumed that all and onlyauxiliary verbs are specified as [AUX �], and that the grammar of the NICEproperties makes reference to this specification. Sag’s (to appear) proposalis different. He shifts the burden of the [AUX �] specifications to the NICEconstructions. Although nonauxiliary verbs are all specified as [AUX �](as in previous accounts), the lexical entries for the auxiliary verbs in thisanalysis are unspecified for the feature AUX, and hence may take part inauxiliary and nonauxiliary constructions alike. This subtle reinterpretationand redistribution of the feature AUX holds the key to understanding theproperties of the exceptional auxiliary do, as we shall see.

Thus the word will has a basic lexical entry like (150):

(150) will��

HEAD

��verb

POL �

FORM fin

��

COMPS

�VPh

FORM basei�

SUBJ h NP[nom] i

��


This verb can head a simple finite VP, and when it does it will be required(according to a constraint we will present in a moment) to be specified as[AUX �].

In section 5.1, we sketched Sag’s analysis of polarized finite auxiliaries.For concreteness, let us think of these forms as derived by lexical rules thatapply to finite auxiliary verbs. Each of these three lexical rules (let us referto them as Adverb Addition, Contraction, and Focus Introduction) requiresthat their input be compatible with the specification [AUX �] and guaran-tees that the rule output is [POL �]. To recapitulate, the grammar producesthree polarized forms derived from (150)—those sketched in (151):

(151) a. will��

HEAD

��

verb

AUX �

POL �

FORM fin

��

COMPS

�ADVpol ,

VPhFORM base

i�

SUBJ h NP[nom] i

��

b. won’t��

HEAD

��

verb

AUX �

POL �

FORM fin

��

COMPS

�VPh

FORM basei�

SUBJ h NP[nom] i

��

c. WILL��

HEAD

��

verb

AUX �

POL �

FORM fin

��

COMPS

�VPh

FORM basei�

SUBJ h NP[nom] i

��


The entries in (151)a–c correspond to the modals used in (152)a–c, respec-tively:

(152) a. Sandy will not go.

b. Sandy won’t go.

c. Sandy WILL go.

The way Sag proposes to correlate polarization and ‘auxiliarity’ is tomodify slightly the constraint on the fin-vp construction presented in sec-tion 2.2 above. The following revision provides a simple correlation of thetwo properties:

(153) fin-vp (final version):

[ ] � H

��HEAD

��verb

POL �

AUX �

��

COMPS A

��, A

What (153) says is that an instance of the fin-vp construction must be[AUX �] and [POL �] or else [AUX �] and [POL �]. Since auxiliary verbsare generally unspecified for the feature AUX (see above), they may appearin both polarized and unpolarized instances of this construction.

The auxiliary verb do is exceptional in that it is lexically specified as[AUX �]. This means that nonpolar finite forms of do cannot appear in thefin-vp construction. Polarized forms of do can head fin-vp phrases, how-ever, as their POL and AUX values are compatible. This treatment thusprovides a straightforward account of well-known contrasts like the fol-lowing:

(154) a. *Tracy dıd leave.

b. Tracy DID leave.

c. Tracy didn’t leave.

d. Tracy did not leave.

e. Tracy did too/so leave.

As noted earlier, if we further require that ADVpol -selecting forms be unfo-cussed, then we also obtain an account of the deviance of examples like(155):


(155) *Kim DID not leave.

As far as do is concerned, there is nothing more than this that a descrip-tively adequate grammar must guarantee. There are systematic semanticrelations between pairs like (156)a,b, of course:

(156) a. Kim likes vindaloo.

b. Does Kim like vindaloo?

But like is [AUX �] and cannot enter into the inversion construction dis-cussed below and does is semantically vacuous (modulo tense) and CAN

appear in inversion constructions. These facts, taken together, are sufficientto guarantee that to utter (156)b is to question the proposition asserted byuttering (156)a. No more direct relation between these two sentences needsto be established by grammar. Our account must only ensure that all andonly the NICE constructions allow [AUX �] verbs.��

7.2. Subject-Aux Inversion Constructions

In the Introduction, we noted that similarities between the constraints onnegation and question inversion in both English and French have been usedto argue for head-movement analyses:

(157) a. *Likes Lou Sandy?

b. Aime-t-il Stacey?

If interrogative sentences are generated by the movement of a verb intoInfl, the contrast in (157) appears to fall out naturally within a systemwhere there are no restrictions on which French verbs move to Infl. Andgiven that in English verbs like have and be may move to Infl, the head-movement

analysis seems to provide a unified account of negation and inversion.However, matters are not quite this simple. There are certain exceptions

that present problems for the analysis of inverted interrogatives via move-ment transformation. Observe the following contrast (due to Joe Emonds,as cited by Chomsky (1981: 209)):

�� This solves the problem that faced earlier G/HPSG analyses that treated AUX spec-ifications as a lexical property. Since auxiliary verbs must be allowed in the fin-vp con-struction and all auxiliary verbs were [AUX �], it followed that that construction had tobe consistent with [AUX �]. Hence that construction was consistent with unfocussed doas well, incorrectly allowing examples like (154)a. This consequence is avoided in ouranalysis because auxiliary verbs other than do are unspecified for AUX, but the fin-vp con-struction requires that the verbal head be [AUX �]. This correctly allows all auxiliary verbsexcept do to head a simple VP.


(158) a. I shall go downtown.

b. Shall I go downtown?

Here there is a semantic difference between the auxiliary verb shall in(158)a and the one in (158)b: the former conveys futurity whereas the latterhas a deontic sense.

Similarly, the following pair exhibits a scope difference (examples dueto John Payne, as cited by Gazdar et al. (1985: 64)):

(159) a. Kim mightn’t go.

b. Mightn’t Kim go?

In (159)a, the modal has scope over the negation (‘It is possible that Kimmight not go.’), whereas in (159)b, only the reverse scope is possible (‘Isit not the case that possibly Kim will go?’).

Further, there are inflected forms that occur only in inversion construc-tions, e.g. the first person singular negative contracted form of the copulaillustrated in (160) (examples from Langendoen 1970; see also Hudson1977 and Gazdar et al. 1982.):

(160) a. *I aren’t going.

b. Aren’t I going?

As far as we are aware, no treatment of these observations has ever beenoffered in transformational terms. It is somewhat unclear how to restrict aparticular inflected form like aren’t so that it will occur only in the structurethat results after movement has applied, or how to restrict scope assign-ment rules in the relevant way. However, in the lexicalist analysis sketchedabove, contrasts like those just noted find a more comfortable home.

As Fillmore (1999) argues at length, the construction type subject-auxiliary-inversion (sai-ph) has numerous subtypes in English.�� Theseare usually left untreated in discussions of the English auxiliary system(in fact, they are usually relegated to the ‘marked periphery’), but theyare well within the descriptive scope of the framework assumed here. Thisconstruction type is constrained as shown in (161):

�� These subtypes include exclamatives (Am I tired!), auxiliary conditions (Were they toagree to that,...), matrix polar interrogatives (Is Kim tired?), and others (May your hair fallout at midnight!). See Newmeyer 1998, pp. 46–49 and Fillmore 1999.


(161) sai-ph:

hSUBJ h i

i� H

��

word

INV �

AUX �

SUBJ h i

COMPS h � ,..., n i

��, , � , � � � , n

In this kind of phrase, which must be headed by an inverted ([INV �]),finite, auxiliary verb, elements are ‘cancelled’ from both head daughter’sSUBJ list and its COMPS list. Again, further constraints on such phrasesare consequences of the GHFP and other general constraints—they do nothave to be stipulated. What does need to be stipulated, following Fillmoreand Ginzburg and Sag, is that there are a number of subtypes of sai-ph, allof which inherit the constraints shown in (161).

It then follows that all auxiliary verbs compatible with the requirementsin (161), including unfocussed do, may appear in matrix polar interroga-tives, in ‘negative adverb preposing’ constructions, in matrix wh-interrogatives, exclamative constructions, and so forth:

(162) a. Did Sandy get the job?

b. Never did they play better!

c. When did Pat say that?

d. Boy, do they stink!

The basic structure for an inverted polar interrogative is sketched in (163):

(163) S��HEAD �

SUBJ h i

COMPS h i

��

V��

word

HEAD �

�INV �

AUX �

�

SUBJ h i

COMPS h � i

��

NP �VP

did Sandy get the job


The feature INV allows an analysis of the lexical idiosyncrasy noted ear-lier. The [INV �] finite auxiliary better, for example, is incompatible withthe requirements of (161) and appears in none of the subtypes of sai-ph.The first-person contracted form aren’t, a form of shall conveying futurity,and a form of mustn’t assigned a ‘not-possible’ reading can be lexicallyspecified as [INV �]. Hence they can appear in inversion constructions,but not in the declarative construction decl-hs-cl illustrated in section 2.2above. In addition, if we assume that the lexicons of vernacular Americanvarieties further constrain the not-selecting auxiliaries to be [INV �], thenwe can account for Bresnan’s (2000) observation that examples like (164)do not occur in these varieties:

(164) a. %Will they not stop singing?

b. %Have they not been to Prague?

It is not clear how lexical peculiarities of this kind can be treated in ananalysis based on head movement.

Similar questions arise about the putative generalization underlying headmovement in the grammar of French. If head movement in French nega-tion and interrogative constructions were a unified phenomenon, then wemight expect that all finite verbs in French should be able to move to C,just as they move to Tense in negative constructions.

But there is idiosyncrasy regarding inversion that seems hard to recon-cile with verb-movement analyses. First, if finite verbs in French were tomove across the subject to form a question, then we should expect theinversion process to be indifferent to the lexical nature of the subject.But this is not the case—inversion is impossible when the subject is non-pronominal, as shown in (165).

(165) a. Jean a perdu son livre.

John has lost his book

b. *A Jean perdu son livre?

Rather, such a question is asked in one of the following ways:��

(166) a. Jean a-t-il perdu son livre?

�� In sentences like (166)a, there are two subjects, the non-pronominal subject NP andthe coindexed weak form pronominal subject. This has caused considerable difficulty formovement-based analyses, where the pronominal subject is taken to be an independentsyntactic element. See Rizzi and Roberts 1989, Drijkoningen 1990, and de Wind 1994 forfurther discussion.


b. A-t-il perdu son livre?

c. Est-ce que Jean a perdu son livre?

Why should a syntactic process like head movement, a specific instantia-tion of Move �, be sensitive to the pronominality of the subject?

And, second, even if we find a way to explain why movement requiresa pronominal subject, the existence of lexical idiosyncrasy again drivesa wedge between negation and question inversion. As noted by Miller(1991), for most verbs there is no acceptable form for the inverted firstperson singular pronominal affix je:

(167) a. Sors-tu?

‘Are you going out?’

b. *Sors-je?

(Putatively) ‘Am I going out?’

Again this exceptionality arises only with respect to true inversion con-structions, not negation, as the following example demonstrates:

(168) Je (ne) sors pas.

‘I am not going out.’

Thus, even in a head-movement analysis, these forms cannot be treatedas exceptions to head movement. The grammar must have some mecha-nism for dealing with elements that are exceptions only to true inversion,a family of constructions that does not include sentential negation.

A third, related issue arises with respect to present participle construc-tions:

(169) a. N’etant pas interesse par la syntaxe, Pierre fait de la phonologie.

‘Being not interested in syntax, Peter works in phonology.’

b. *Ne pas etant interesse par la syntaxe, Pierre fait de la phonolo-gie.

Present participles in French, unlike infinitives, must precede the negationpas. To account for this fact, Pollock (1989) suggests that present partici-ples in French are finite and that they are hence forced to undergo move-ment in order to satisfy his quantification theory.

This last proposal raises questions about the semantic nature of presentparticiples in relation to the standard interpretations provided by tense


operators. Why, for example, should present participles, but not past par-ticiples, have a semantic analysis that makes use of variable binding? Leav-ing these concerns aside, however, the verb movement analysis remainsdeficient. Since in the verb movement analysis all verbs that precede thenegation also precede the subject in questions, it should follow that presentparticiples also precede the subject in questions. But of course this is notthe case in French—present participles do not give rise to interrogativesentences. In our lexicalist analysis of these French phenomena, we treattensed and pres(ent)-part(iciple) as two subtypes of finite. As argued insection 5.1, all finite forms may select pas as a complement, whereas theinverted ([INV �]) clauses are headed only by verbs specified as [FORMtensed]. These ([INV �]) forms are either lexically restricted to require apronoun subject or else (following Miller 1991), these forms take no sub-ject, but realize a pronominal element as an inflectional affix. Under eitherapproach, the impossible occurrences (e.g.*Sors-je?) can be treated as lex-ical gaps.

8. CONCLUSION

In this paper, we have developed a picture of the comparative grammarof English and French that is radically different from the one offered byhead-movement analyses. On our account, the two languages share numer-ous constructions. In particular, the French and English decl-hd-su-cl andmod-hd constructions are virtually identical. Indeed, the two systems ofnonfinite negation ([not/ne-pas VP[nonfin]]) are almost indistinguishable.Certain differences in adverbial position (e.g. the difference between (ne-)jamais and never, occurring as true modifiers) are purely lexical in nature.There are also lexical differences that make inversion constructions subtlydifferent in the two languages, despite a common constructional core.

The centerpiece of our study, however, has been the head-complementstructures. We have argued, following Abeille and Godard, that finite nega-tion in French should be analyzed in terms of a general pattern that per-mits finite verbs to select a negative adverb as a complement. In English,by contrast, it is only finite forms of auxiliary verbs that may select for apolar adverb (not, too or so). Moreover, we have seen that these Englishforms are part of a larger system of polarized auxiliaries, including con-tracted and focussed finite forms. It is only when we look at this systemas a whole that we are able to explicate both the lexical exceptions and thegeneralizations that govern auxiliary-related phenomena, including suchlong-standing puzzles as the distribution of auxiliary do.


We have shown that a (strongly) lexicalist, surface-based approach togrammar can explain the critical phenomena in each language (includ-ing massive lexical idiosyncrasy) and also provide a principled basis forexplaining the syntactic differences between the two languages. We haveachieved this result in terms of theoretical foundations quite different fromthose assumed in most current work. It is of course possible that some-one will someday develop a grammar that uses verb movement and func-tional projections to account for the numerous phenomena we have ana-lyzed here. We are not arguing that this is impossible; but it appears thatany attempt to modify the movement analysis to accommodate these datawill involve introducing devices that deal with lexical sensitivity to derivedstructure. Executing this program successfully will lead, we believe, to ahighly surface-sensitive account similar to our own, where head movementhas no role to play.

REFERENCES

Abeille, Anne and Daniele Godard. 1994a. The Complementation of TenseAuxiliaries. In Raul Aranovich, William Byrne, Susanne Preuss, andMartha Senturia (eds.), Proceedings of the West Coast Conference onFormal Linguistics 13, 157–172. Stanford: SLA, CSLI Publications.

Abeille, Anne and Daniele Godard. 1994b. The Syntax of French Auxil-iaries. Ms. UFRL and CNRS, Universite Paris 7.

Abeille, Anne and Daniele Godard. 1996. La complementation des auxil-iares Francais. Langages 122: 32–61.

Abeille, Anne and Daniele Godard. 1997. The Syntax of French NegativeAdverbs. In Paul Hirschbuhler and F. Marineau (eds.) Negation andPolarity: Syntax and Semantics, 1–17. Amsterdam: John Benjamins.

Abeille, Anne and Daniele Godard. 1998. A Lexical Approach to Quanti-fier Floating in French. In A. Kathol, J.-P. Koenig, and G. Webelhuth(eds.), Lexical and Construction Aspects of Linguistic Explanation.Stanford: CSLI Publications.

Abeille, Anne, and Daniele Godard. 2000. French Word Order and LexicalWeight. In R. Borsley, ed., Syntax and Semantics Volume 32: TheNature and Function of Syntactic Categories, 325–360. San Diegoand London: Academic Press.

Ackerman, Farrell and Gert Webelhuth. 1998. A Theory of Predicates.Stanford: CSLI Publications.


Auger, Julie. 1994. Pronominal Clitics in Quebec Colloquial French: Amorphological analysis. Doctoral dissertation, University of Pennsyl-vania. Published as IRCS Report 94-29.

Baker, C. L. 1971. Stress Level and Auxiliary Behavior in English. Lin-guistic Inquiry 2: 167–181.

Baker, C. L. 1989. English Syntax, Cambridge, MA: MIT Press.

Baker, C. L. 1991. The Syntax of English Not: The Limits of Core Gram-mar. Linguistic Inquiry 22: 387–429.

Belletti, Adriana. 1990. Generalized Verb Movement. Torino: Rosenbergand Sellier.

Borsley, Robert. 1994. In Defense of Coordinate Structures. LinguisticAnalysis 24: 218–246.

Bouchard, Denis. 1997. The Syntax of Sentential Negation in French andEnglish. In P. Hirschbuhler and F. Marineau (eds.) Negation and Polar-ity: Syntax and Semantics, 29–52. Amsterdam: John Benjamins.

Bouma, Gosse, Frank van Eynde, and Daniel Flickinger. 2000. Constraint-Based Lexica. In F. V. Eynde and D. Gibbon, eds., Lexicon Develop-ment for Speech and Language Processing, pages 43–75. Dordrecht:Kluwer.

Bouma, Gosse and Gertjan Van Noord. 1994. Adjuncts and the Processingof Lexical Rules. Proceedings of COLING 1994, Kyoto.

Bouma, Gosse, Rob Malouf, and Ivan Sag. 2001. Satisfying Constraints onExtraction and Adjunction. Natural Language and Linguistic Theory.19.1: 1–65.

Bresnan, Joan. 1976. The Form and Functioning of Transformations. Lin-guistic Inquiry 7: 3–40.

Bresnan, Joan. 2000. Optimal Syntax. In Optimality Theory: Phonolo-gy, Syntax, and Acquisition, edited by Joost Dekkers, Frank van derLeeuw and Jeroen van de Weijer. Pp. 334–385. Oxford: Oxford Uni-versity Press.

Bresnan, Joan and Sam A. Mchombo. 1995. The Lexical Integrity Princi-ple: Evidence from Bantu. Natural Language and Linguistic Theory,13.2: 181–254.

Carpenter, Bob. 1992. Categorial Grammars, Lexical Rules, and the EnglishPredicative. In R. Levine (ed.), Formal Grammar: Theory and Imple-mentation. New York and Oxford: Oxford University Press.

Chomsky, Noam. 1981. Lectures on Government and Binding. Dordrecht:Foris.


Chomsky, Noam. 1991. Some Notes on Economy of Derivation and Rep-resentation. In R. Friedin (ed.), Principles and Parameters in Com-parative Grammar, 417–454. Cambridge, MA: MIT Press.

Chomsky, Noam. 1993. A Minimalist Program for Linguistic Theory. InKenneth Hale and Samuel Keyser (eds.), The View from Building 20,1–52. Cambridge, MA: MIT Press. Reprinted in Chomsky 1995.

Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: MITPress.

Cinque, Cuglielmo. 1999. Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford: Oxford University Press.

Copestake, Ann, and Edward Briscoe. 1992. Lexical rules in a unifica-tion based framework. In J. Pustejovsky and S. Bergler (eds.), LexicalSemantics and Knowledge Representation, Lecture Notes in ArtificialIntelligence Number 627, pages 101–119. Berlin: Springer-Verlag.

Culicover, Peter W. 1991. Polarity, Inversion, and Focus in English. ESCOL’91 :46–68.

de Wind, Maarten. 1993. Against V-To-C in French Complex Inversion.In Michael Mazzola (ed.) Issues and Theory in Romance Linguistics,271–283. Washington, D.C.: Georgetown University Press

Di Sciullo, Anna M. and Edwin Williams. 1987. On the Definition of Word.Cambridge, MA: MIT Press.

Dowty, David. 1988. Type Raising, Functional Composition, and Non-Constituent Conjunction. In Richard Oehrle, Emmon Bach andDeirdre Wheeler (eds.), Categorial Grammars and Natural LanguageStructures, 153–198. Dordrecht: D. Reidel Publishing.

Drijkoningen, Frank. 1990. Functional Heads and the Unification of FrenchWord Order. Probus 2.3: 291–320.

Emonds, Joseph. 1978. The Verbal Complex V�-V in French. LinguisticInquiry 9: 151–175.

Ernst, Thomas. 1992. The Phrase Structure of English Negation. The Lin-guistic Review, 9: 109–144.

Ernst, Thomas. 1998. The Scopal Basis of Adverb Licensing. In Pius N.Tamanji and Kiyomi Kusumoto (eds.), Proceedings of the North EastLinguistic Society 28, 127–142. Amherst: GLSA.

Fillmore, Charles. 1999. Inversion and Constructional Inheritance. In G.Webelhuth, J.P Koenig, and A. Kathol (eds.), Lexical and Construc-tional Aspects of Linguistics Explanation, 113–128. Stanford: CSLIPublications.


Gazdar, Gerald, Ewan Klein, Geoffrey K. Pullum, and Ivan Sag. 1985.Generalized Phrase Structure Grammar. Oxford: Blackwell’s andCambridge, MA: Harvard University Press.

Gazdar, Gerald, Geoffrey K. Pullum, and Ivan Sag. 1982. Auxiliaries andRelated Phenomena in a Restrictive Theory of Grammar. Language58: 591–638.

Ginzburg, Jonathan, and Ivan A. Sag. 2000. English Interrogative Con-structions. Stanford: CSLI Publications.

Grimshaw, Jane. 1997. Projections, Heads, and Optimality. LinguisticInquiry 28.3: 373–422.

Haegeman, Liliane. 1997. Introduction: On the Interaction of Theory andDescription in Syntax. In L. Haegeman (ed.), The New ComparativeSyntax. London and New York: Longman.

Hirschbuhler, Paul and Marie Labelle. 1994. Changes in Verb Position inFrench Negative Infinitival Clauses. Language Variation and Change6: 149–178.

Hudson, Richard. 1976. Arguments for a Non-transformational Grammar.Chicago: University of Chicago Press.

Hudson, Richard. 1977. The Power of Morphological Rules. Lingua 42:73–89.

Iatridou, Sabine. 1990. About AgrP. Linguistic Inquiry 21: 551-577.

Jackendoff, Ray. 1972. Semantic Interpretation in Generative Grammar.Cambridge, MA: MIT Press.

Johnson, David, and Shalom Lappin. 1999. Local Constraints vs. Econo-my. Stanford: CSLI Publications.

Kasper, Robert. 1994. Adjuncts in the Mittelefeld. In John Nerbonne, KlausNetter, and Carl Pollard (eds.), German in Head-Driven Phrase Struc-ture Grammar, 39–70. Stanford: CSLI Publications.

Kayne, Richard. 1991. Romance Clitics, Verb Movement, and PRO. Lin-guistic Inquiry, 22: 647–686.

Kayne, Richard. 2000. Parameters and Universals. Oxford: Oxford Uni-versity Press.

Kim, Jong-Bok. 1995. English Negation from a Non-Derivational Perspec-tive. In Jocelyn Ahlers et al. (eds.), Proceedings of the 21st AnnualMeeting of the Berkeley Linguistics Society, 186-197. Berkeley, CA:Berkeley Linguistic Society.

Kim, Jong-Bok. 2000. The Grammar of Negation: A Constraint-BasedApproach. Stanford: CSLI Publications.


Kim, Jong-Bok, and Ivan A. Sag. 1995. The Parametric Variation of Englishand French Negation. In Jose Camacho, Lina Choueiri, and MakiWatanabe (eds), Proceedings of the Fourteenth West Coast Confer-ence on Formal Linguistics (WCCFL), 303-317. Stanford: SLA CSLIPublications.

Koenig, Jean-Pierre. 1999. Lexical Relations. Stanford: CSLI Publications.

Laka, Itziar. 1990. Negation in Syntax: On the Nature of Functional Cate-gories and Projections. Doctoral Dissertation, MIT.

Langendoen, D. Terence. 1970. Essentials of English Grammar. New York,Holt, Rinehart, and Winston.

Lasnik, Howard. 1995. Verbal Morphology: Syntactic Structures Meets theMinimalist Program. In H. Campos and P. Kempchinsky (eds.), Evo-lution and Revolution in Linguistic Theory: Essays in Honor of Char-los Otero, 251–75.

Lasnik, Howard. 1999. Minimalist Analysis. Malden, MA: Blackwell Pub-lishers Inc.

Lasnik, Howard (with Marcela Depiante and Arthur Stepanov). 2000. Syn-tactic Structures Revisited. Cambridge, MA: MIT Press.

Lascarides, Alex and Ann Copestake. 1999. Default representation inconstraint-based frameworks. Computational Linguistics 25:1, 55-105.

Lopez, Luis. 1994. The Syntactic Licensing of VP-Ellipsis: A ComparativeStudy of Spanish and English. In Michael L. Mazzola (ed.), Issuesand Theory in Romance Linguistics. Washington: Georgetown Uni-versity Press.

Manning, Christopher, Ivan Sag, and Masayo Iida. 2000. The LexicalIntegrity of Japanese Causatives. In Robert Levine and Georgia Green,(eds.), Readings in Modern Phrase Structure Grammar, 39–79.

McConnell-Ginet, Sally. 1982. Adverbs and Logical Form. Language 58:144–184.

Meurers, Detmar Walt. 2000. Lexical Generalizations in the Syntax of Ger-man Non-Finite Constructions. Ph.D. thesis, Universitat Tubingen.Published as Arbeitspapiere des SFB 340, Number 145. UniversitatTubingen.

Miller, Philip H. 1991. Clitics and Constituents in Phrase Structure Gram-mar. Doctoral Dissertation, Utrecht University.

Miller, Philip H. and Ivan A. Sag. 1997. French Clitic Movement WithoutClitics or Movement. Natural Language and Linguistic Theory. 15:573–639.


Newmeyer, Frederic. 1998. Language Form and Language Function. Cam-bridge, MA: MIT Press.

Partee, Barbara and Mats Rooth. 1983. Generalized Conjunction and TypeAmbiguity. In R. Bauerle, C. Schwarze, and A. von Stechow (eds.),Meaning, Use, and Interpretation of Language, 361–366. Berlin: Wal-ter de Gruyter.

Pollard, Carl and Ivan Sag. 1994. Head-Driven Phrase Structure Gram-mar. Chicago: University of Chicago Press and Stanford: CSLI Pub-lications.

Pollock, Jean-Yves. 1989. Verb Movement, Universal Grammar, and theStructure of IP. Linguistic Inquiry 20: 365–424.

Pollock, Jean-Yves . 1994. Checking Theory and Bare Verbs. In Cinque,Gulielmo et al. (eds.), Paths towards Universal Grammar, 293–310.

Pollock, Jean-Yves. 1997a. Notes on Clause Structure. In L. Haegeman(ed.), Elements of Grammar. Dordrecht: Kluwer Academic Publish-ers.

Pollock, Jean-Yves. 1997b. Language et Cognition. Paris: Presses Univer-sitaires de France.

Pullum, Geoffrey K. 1982. Syncategorematicity and English InfinitivalTo’. Glossa 16:181–215.

Przepiorkowski, Adam. 1999a. Case Assignment and the Complement/Adjunct Dichotomy: A non-configurational constraint-based approach.Ph.D. thesis, Universitat Tubingen.

Przepiorkowski, Adam. 1999b. Adjuncts as complements: Evidence fromcase assignment. In Andreas Kathol, Jean-Pierre Koenig, and GertWebelhuth (eds.) Lexical and Constructional Aspects of LinguisticExplanation. Stanford: CSLI Publications.

Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech and Jan Svartvik.1985. A Comprehensive Grammar of the English Language. London:Longman.

Rizzi, Luigi. 1990. Relativized Minimality. Cambridge, MA: MIT Press.

Rizzi, Luigi and Roberts Ian. 1989. Complex Inversion in French, Probus1, 1–30.

Roberts, Ian G. 1994. Two Types of Head Movement in Romance. InDavid Lightfoot and Norbert Horstein (eds.), Verb Movement, 207–242. Cambridge: Cambridge University Press.

Sag, Ivan A. 1976. Deletion and Logical Form. Doctoral Dissertation. MIT.[published by Garland Publishing, New York. 1980].


Sag, Ivan A. 2000. Another Argument Against Wh-Trace. Jorge HankamerWebfest. http://ling.ucsc.edu/Jorge/

Sag, Ivan A. ms. Without a Trace. Unpublished manuscript, Stanford Uni-versity.

Sag, Ivan A. to appear. Rules and Exceptions in the English Auxiliary Sys-tem. Journal of Linguistics.

Sag, Ivan A. and Janet Fodor. 1994. Extraction Without Traces. In RaulAranovich, William Byrne, Susanne Preuss, and Martha Senturia (eds.),Proceedings of the West Coast Conference on Formal Linguistics 13,365–384. Stanford: Stanford Linguistic Association.

Sag, Ivan A. and Tom Wasow. 1999. Syntactic Theory: A Formal Approach.Stanford: CSLI Publications.

Scalise, Sergio. 1984. Generative Morphology. Dordrecht, Holland andCinnaminson, N.J.: Foris Publications.

Vikner, Sten. 1997. V to I Movement and Inflection for Person in AllTenses. In L. Haegeman (ed.), The New Comparative Syntax. Londonand New York: Longman.

Warner, Anthony R. 1993. English Auxiliaries: Structure and History. Cam-bridge: Cambridge University Press.

Warner, Anthony R. 2000. English Auxiliaries Without Lexical Rules. InR. Borsley, ed., Syntax and Semantics Volume 32: The Nature andFunction of Syntactic Categories, 167–220. San Diego and London:Academic Press.

Wechsler, Steve and Yae-Sheik Lee. 1996. The Domain of Direct CaseAssignment. Natural Language and Linguistic Theory 14.3, 629-664.

Williams, Edwin. 1994. Thematic Structure in Syntax. Cambridge, MA:MIT Press.

Wood, Mary McGee. 1993. Categorial Grammars. London: Routledge.

Zagona, Karen. 1988. Proper Government of Antecedentless VP in Englishand Spanish. Natural Language and Linguistic Theory 6: 95–128.

Zanuttini, Raffaella. 1991. Syntactic Properties of Sentential Negation: AComparative Studies of Romance Languages. Doctoral Dissertation.University of Pennsylvania.

Zanuttini, Raffaella. 1997. Negation and verb movement. In L. Hageman(ed.), The New Comparative Syntax, 214–245. London and New York:Longman.

Zwicky, Arnold M. and Geoffery K. Pullum. 1983. Cliticization vs. Inflec-tion: English n’t. Language 59: 502–513.


First received 8/28/00Revision received 1/09/01Final revision received 7/22/01

Jong-Bok KimSchool of English, Kyung Hee UniveristySeoul, 130-701 [email protected]

Ivan A. SagDept. of Linguistics, Stanford UniversityStanford, CA 94305 [email protected]

Documents

1. I The similarities and systematic differences between English …web.khu.ac.kr/~jongbok/research/final-papers/kim-sag... · 2015-03-27 · our proposed analysis. 2.1.1. Lexicon